Episode#126| Biomaterials Frontier for Medical 3D Printing (Virtual Event Recording)

The real frontier in medical 3D printing isn't the printer — it's the material. In this 3DHEALS panel, five researchers, founders, and industry leaders map the shift from printing structure to printing function, and from lab-stage novelty toward regulated, investment-grade material systems spanning soft bioresins, bioactive ceramics, and titanium lattices.

Fabian Trumper (Arrakis Bio) argues that the best biomaterials are the ones our bodies already make, and explains how his team bio-manufactures True Human Collagen at industrial scale. Prof. Kamal Choudhary (Johns Hopkins) shows why AI is reshaping materials discovery — a "ChatGPT for materials scientists" — and why so much of the field still isn't reproducible. Dr. Andrew Weems (Resilient Medical) shares how a 3D-printed scaffold could finally modernize lumpectomy surgery for breast cancer patients. Ebrahim Yarali (MERLN, Maastricht University) reveals how geometry alone — architected "meta-biomaterials" and 4D printing — can steer stem cells toward bone or cartilage. And Dr. Scott Taylor (Poly-Med) breaks down absorbable polymers that dissolve into the body once healing is done. Moderated by Craig Rosenblum, President of Himed.

In this episode:

• Human vs. synthetic biomaterials — and what cells actually "notice"
• AI-designed materials and the reproducibility crisis
• Designing devices for the surgery we already do
• 4D printing and geometry as a biological signal
• A live debate: can AI and lab models replace animal testing?

Whether you're a researcher, founder, or investor, this is a fast, opinionated tour of where medical materials are heading next.

🎥 Watch the full session and explore more 3DHEALS events: https://3dheals.com/biomaterials-frontier/

00:00:00 - Welcome And 3D Heals Missions
00:01:58 - Event Setup And Sponsor Context
00:04:08 - Absorbable Polymers And Degradation Design
00:10:07 - Pore Size And Surface Area Effects
00:12:40 - True Human Collagen For Regeneration
00:23:08 - Collagen Q&A On Regulation
00:28:58 - AI Tools For Materials Discovery
00:41:35 - AI Limits Data And Loop Closure
00:50:31 - Lumpectomy Scaffolds And Go To Market
01:01:47 - Academia Versus Startup Mindset
01:10:13 - Geometry Driven Bone Regeneration Scaffolds
01:24:11 - Acoustic Responsive And 4D Biomaterials
01:31:08 - Panel Debate On Animal Testing
01:37:39 - AI Hype Real Impact And Wish List
01:46:51 - Final Takeaways And Closing

Transcript

Copy Transcript Copied

Welcome And 3D Heals Missions

SPEAKER_02 0:00

Hello, hello, good morning everyone. My name is Jenny Chen, CEO and founder of 3D Heals. We have been doing these kind of events since 2020, the pandemic, but 3D Heels was funded in 2017, so almost 10 years. And we have three missions. One, education to put on events like this for people to learn more about the technology and not just the superficial things that looks pretty, but really functional and critical components and devices for healthcare. And number two is networking. So for virtual events, it's a lot harder to make friends, but you can still do that. And one way to do that is share your social media link. If you have LinkedIn and you want to connect professionally here, feel free to share or email us or just tell us who you are and what you want to learn or or need for your career at the moment. Number three is Pitch 3D program. We have our early stage startup program that we help these startups funding fund institutional fundings. So we're connected with more than 30 institutional investors, which means VCs, corporate VCs, and we help early stage investor uh startup to connect with these investors. And the program is free. And if you're interested, feel free to contact me directly. But without further ado, I'd like to introduce the topic today, which I think is probably one of the most important topics in medical 3D printing, which is material innovation. And all the experts here today are from all aspects of this big topic, material science. So hopefully we have, you know, a little bit of taste of what things are going on on the frontier side and at the edge of this field right now, especially for 3D printing. We also have an excellent moderator, Craig Rosenblum. He is the president of Hymet, and he himself is extremely knowledgeable with both material science and also post-processing for 3D printing, but I'll let him take

Event Setup And Sponsor Context

SPEAKER_02 1:58

over now.

SPEAKER_00 1:58

Excellent. Well, thank you, Jenny. Thank you to the members of our panel and thank you for the attendees, which are slowly but surely trickling in. We have an exceptional event today. First off, I'd like to congratulate and thank Jenny for all of her hard work in pulling this event together. I agree. Obviously, we're both biased, but I think the Biomaterials Frontier event is really the premium uh event that 3D Heels hosts. For one, we have panelists that are located all around the world. Uh so a lot of people are up very early in the morning or staying up very late at night for this particular event. As Jenny had indicated, the value of this event really is to engage, and some of the best conversations take place after the panel discussion. So certainly, while our speakers are presenting, feel free to connect with folks, whether it be in the chat or separately through LinkedIn. And I think we would all be happy to engage in separate conversations. Let me introduce myself to begin our conversation, and also I'll introduce our co-sponsor and companies, both HiMed as well as PolyMed, for today's event. My name again is Craig Rosenbloom. I'm president of HiMed. HiMed is a biomaterials company. I see Scott has just joined also, so I'll allow Scott separately to introduce Polymed. But HiMed is a biomaterials company, and we specialize in the manufacture of a variety of biomaterials, predominantly calcium phosphates, which are used to enhance the biological surfaces of medical implants, and they're also used for orthobiologic medical devices. Naturally, with the continued technology emergence with additive manufacturing, our company is heavily involved both in using apatitic abrasive, a hydroxyapatite-based blast media for post-processing as well as for 3D printing of bioceramics in our newly established Bioceramic Center of Excellence. I'll drop some links in the chat, but I wanted to introduce Hymed as our first co-sponsor for this event, and then Scott, in the perfect nick of time, uh will introduce himself and uh talk about his company as well.

Absorbable Polymers And Degradation Design

SPEAKER_03 4:08

Yeah, absolutely. Well, it's it's great to uh to finally uh be able to join. I am, as Craig said, and and uh we go back uh to to some prior meetings together. Um, you know, we're we're honored to be a part of this as co-sponsors of this event. My name is Scott Taylor. I'm I'm director of product engineering at PolyMed. And you know, what we do we're as I'm trying to uh a slide, but I don't believe we're gonna be able to do that. That's fine. So you know, what what we do at Polymed is uh, you know, we uh we try to advance the state of of medical devices through the development and manufacture of absorbable polymer systems, through the development of new polymer formats, uh, as well as the conversion of that into semi-finished and finishing medical devices. So we are a uh a C DMO with extensive capabilities of uh of uh material design all the way through processes and ultimately manufacturing. And we believe that really this starts with the question of um of performance, right? What performance do we expect absorbable medical devices to do? Because a lot of times, you know, these do have traditional mechanical functionality as a requirement in the early stages of performance. And then over time, these materials all degrade into naturally occurring byproducts that the that the body can process and eliminate from the system. So you're left with just the healed tissues, the uh the residual, you know, healthy, normal tissues that are left behind as these as these devices degrade. And and really one of the things that we're focused on is turning these these traditionally simple devices, right? The earliest versions of absorbable medical devices were sutures, and and we are converting uh these early mechanical devices into increasingly complex solutions for uses as tissue scaffolding, uh temporary functionality, volume replacement, and really and truly trying to uh encourage uh the use of these types of materials in regenerative medicine devices. So uh, you know, we do this in a number of ways, but for us it always starts with a material. So by smarter polymer design, by smarter polymer selection, uh we are able to more readily impact a positive outcome for wound healing. There's two things specifically that that we're working on, I think that we're very excited to discuss. And one of these we actually spent some time and launched that at Indian West earlier this year. And this is uh this is one of our newer colonal platforms, we call it lactoprene HMX. So this is a high modulus uh absorbable biomaterial, um, and it is a material that we can mill, we can 3D print, and we have we have shown that this that material is safe and compatible with the body for a long-term degradable platform. So we can retain strength in orthopedic type applications for six months or longer with molded or with fibrous versions of these articles. And one of the things that we're very excited about is actually the mechanics of this material. So traditional polylactide, um, you know, we're talking a tensile modulus of of maybe one to two gigapascals. And really, we know in orthopedic applications, bone is closer on the order of maybe eight to twenty gigapascals in stiffness in that tensile modulus. And and what we have shown with the lactoprene HMX through uh through the material and through the processing that we use is we're able to achieve a tensile modulus closer to five gigapascals or larger with very good creep performance, very good surface hardness and and uh and degradation rate. So, you know, we're we're very excited about these materials in terms of screws, plates, anchors, and tacks. You know, the other thing that we are spending a lot of time on is 3D printing, right? That's that's why we are a part of this this 3D heels organization. And in particular, photoset resin is a material that we have developed. It's it's fully degradable and it is intended for VAT photopolarization into very complex and highly tolerant structures that it we have shown by compatibility throughout the degradation cycle. And so, and we've we've shown that in both in vitro as well as through animal studies, through you know, very low reactivity, both in the three-day, so that early inflammatory period, as well as uh, you know, up to you know, a month or two out from degradation. There's very low capsule, very high biocompatibility response throughout the degradation life of these materials. And you know, that in particular with that photoset material, we're we're looking at uh the ability to create very complex tissue scaffolds and void fillers for soft tissue applications. But in general, polymat is much more than just those two materials. We start with polymer, like we said, so we we do synthesize a significant amount of absorbable polymers here. Uh, and then we also convert those through 3D printing, through electrospinning, through traditional fiber and film extrusion, and then convert all of those things into biomedical textiles, such as knitted structures that are ultimately used throughout the body. I don't believe there is a surgical specialty that our materials and our processes have not supported over our 30-year history. And in the end, what Polymed wants to wants to be known as, what we strive to be, is trusted partners that are dedicated to advancing patient care through the development, through the creation of these absorbable biomaterials, components, and implantable medical devices. And we do that through this history of research and development, clinical insights, and then advanced manufacturing that are specifically tailored to these complex materials. With that, I'll I'll turn it back over to uh to

Pore Size And Surface Area Effects

SPEAKER_03 10:07

Jenny. So thank you guys for allowing us to be here today.

SPEAKER_00 10:10

Thank you, Scott. Scott, I have a question for you before we introduce our first panelist. Absolutely. How does the pore size influence that degradation profile? Uh I feel like a critical question that you know probably would be a valuable point to discuss so people have a better understanding of what polymed's contribution is to the conversation.

SPEAKER_03 10:30

Yeah, absolutely. Yeah, so uh in pore size, you know, we think of it in terms of just raw volumetric space, right? And the goal of all of the devices that we make is to elicit a healing response so that the body does not need our materials long term, right? So we're we're uh encouraging healing back to pre-injury state. Uh pore size, as well as that surface area to volume ratio does have influence on the healing kinetics, right? So there is sort of a sweet spot of pore size that can encourage ingrowth, support that protein deposition on the surface on raw materials that can elicit uh neovascularization. And and uh, but generally the higher the surface to volume ratio, the faster degrading these materials are. So we're talking about electrospinning, highly porous, very, very high surface area compared to a traditionally molded part. And so there's just more opportunity to interact in a positive way with the body with higher surface to volume and more porosity.

SPEAKER_00 11:31

Very good. And I see that Corey and Tamara, and I'm sure that there's others from your company that are on the conversation as well. So thanks for sponsoring the event. Yeah, thanks, Craig. All right. Without further ado, I think we're gonna begin our panel discussion, and I see Fabian on the call. Let me introduce Fabian. Fabian is the co-founder, Fabian Trumper is the co-founder and CEO of Arrakis Bio, an Israeli biotech company revolutionizing the field of regenerative human biomaterials with its true human collagen. He's also a three-time founder with over two decades of product technology, business leadership across deep tech and life sciences, including as the VP and pro of products and co-founder of Lightbits Lab, and CEO and co-founder of Buntica. Fabian holds an MBA in marketing strategy from Inseed. And Fabian, without further ado, feel free to take over the floor.

SPEAKER_01 12:29

Thank you, Craig. Let me share some slide material that prepared for this one. See if I can do that. All right. Can you see the slides here? Yes. Okay.

True Human Collagen For Regeneration

SPEAKER_01 12:40

So good morning, Jenny, Craig, and all. Thank you for inviting me to this panel. I'm Fabian Trumper, CEO and co-founder of Araki's Bio. And we grow human biomaterials. And over the next few minutes, I'd like to make it one case that the biomaterials that our body naturally produces are the best ones. And a great deal could change once we recognize that. So any biomaterial has to satisfy two masters at the same time. It has to meet strict mechanical and structural requirements, and it also has to satisfy the biological requirements of the cells and tissue that it would interact with. And almost everything available today wins on one and loses on the other one. So if we zoom out and we look at the landscape of what we have today, the material options, a big picture, we have three categories. We have human biomaterials, human-native biomaterials, we have natural materials but non-human, and we have synthetic biomaterials. So synthetic biomaterials like polymers, for example, which we've heard just now, they print beautifully, they're reproducible, they're affordable, they are they hold their shape. But to a cell they look like plastic. They are mechanically functional, but biologically they are inferior to everything that the cells are familiar with. So to bring back some of the biological function, the default compromise in the industry is to incorporate also natural biomaterials or to use natural biomaterials, and typically from non-human sources, animals or plants. And in these materials work up to a point. You get multiple issues that you inherit with the source of these materials. One is immune interference. You cannot really fool our immune system. These are foreign bodies. So our body reacts to that as a foreign body. They're xenogenic in origin, they're not normally there, and we don't really know how that affects the cells. But if they were better than what we already have in there, then evolution would make itself. We get biological variability in our lot-to-load differences that are unavailable, unavoidable due to the source of the materials coming from different sources. And ultimately, there's also a regulatory outlook that is kind of shifting around the industry and driving us away from animal-derived materials. So, overall, looking at the current state of the art, we get the mechanics that are pretty much solvable. The biology is where we fall short. So, can we do better? And this brings me to the thesis behind Iraqis bio. And we believe that the next generation of biomaterials should be based on human biomaterials. And I'll try to make the case for that. And if you pick one biomaterial to solve first, that would be collagen. It's the most abundant protein in the human extracellular metrics. It's the backbone of almost every tissue that we would want to build. And it is fundamental to almost every regenerative process in our body. So that's the opportunity that we're looking at. And I would say that pretty much the research community agrees with that. If you look at the number of publications on collagen and bioprinting, and this is the chart that we are looking at now, the number of publications from the last 25 years, right? It has grown significantly and continues to accelerate. And compared to any other biomaterial, it is an order of magnitude higher. So pretty much the research community agrees that collagen has a huge potential. But 99% of the collagen used today is not from human origin. So you might wonder what would happen if we used human collagen instead of animal-derived collagen. Does the source really matter so much? Does it matter if it is from human or a rat? So look at the gene expression differences in the chart on the right. This is from a recent research paper that compared human collagen to rat tail collagen. You change the collagen source and you change the gene expression in a significant way. And that, of course, affects also tissue structure and even aging markers, as I will demonstrate later in the presentation. So the ECM makes a difference, that is clear. And the science isn't the only thing that is pushing us this way. The regulatory bodies are also stepping up to meet the biology. Animal-derived and animal tested models are being phased out, at least. That is the outlook for the next few years. And human-relevant models are being pulled in by the regulatory bodies around the world, which is why we expect human biomaterials to follow and not just be a nice to have but a must-have going forward in the next few years. And it's it's not just a research project or a lab experiment. This is getting out of the lab and is already entering commercial products. The first human cornea transplant using 3D printed and lab-grown tissue built on human collagen was successfully performed in a phase one trial last year by a company called Precise Bio, which happens to be just a few miles away from where I sit. So the advantages are clear. Using human collagen, we get biocompatibility, we get cell addhesion, we get high cell loading, we get cell proliferation, we don't get immune interference, and it integrates with the implant site and actively drives tissue regeneration rather than just sit around or be a mechanical support. So are we done? Are we do we have the ideal biomaterial nailed down? So unfortunately, not just yet. There are a few challenges to address. One, the mechanical characteristics, the printability and stability in situ, and finally cost. And I would argue that the first two are largely solved in the recent years, and that the third one is the real story here. So on the mechanical side, the technology toolbox has pretty much caught up in the last few years. Technologies like fresh bioprinting and others let us build very precise, perfusible, and vascularized structures using collagen. And there's a growing number of technologies that also help us tune the mechanical properties of the bioprinted collagen. Things like accelerated gelation, fibrillation, photoinitiated gross-linking, and also multimaterial compositions to reconstruct a wide range of different tissues. And that's why I would argue that these challenges are no longer the main constraint for human collagen adoption. Which leaves us with the final challenge: supply and cost. And that's what we believe is the real barrier that is stopping mass adoption of human collagen in medical application. The industry doesn't know how to biomanufacture human collagen. The only way we know how to get it is to harvest it from donated tissue. And that's a scarce and unscalable resource. So the key for human collagen adoption across the entire field is unlocking the economics and scalability. And that is the problem that Iraqis bio is aiming to solve. And there are three pieces to our technology: an engineered human fiberglass cellite, a bioreactor-based production system, and an efficient purification and formulation process. And put that all together, we have the technology to biofabricate 100% human biomaterials at industrial scale and cost. And you might ask why we use cell culture rather than recombinant systems. And the problem with recombinant methods, which have their place, is that they cannot reproduce the huge complexity of human extracellular biomaterials. And taking collagen as an example, there are more than 10 enzymes involved in producing and secreting native collagen correctly. And recombinant systems lack the biological machinery to reproduce most of them. So if we want genuinely native human collagen, we need to go for human cells to make it. And it's not just about collagen, of course. We are talking about producing the whole human extracellular matrix, collagen, elastin, growth factors, and more. But let's just focus on collagen for now and see where what we get. So what we see you see in this picture is our cell line producing collagen. And what makes our technology economically viable is the transition from the left-hand side where you see untreated cells to the right-hand side where we get 50 times more collagen from the same cells. And that was what makes the cost structure work and the collagen availability an economically viable solution. And it's it's just yielded, the product itself has to be the real thing, of course. So we characterize it properly, and across the panel, our collagen stands up to the test. And more than that, our collagen composition mirrors exactly our nature, natural skin composition. It produces both collagen type 1 and 3 at the exact ratio that you would find in our skin. And that's no surprise, of course, because these are dermal fibroblasts. They simply do what they know how to do best, and that's produce human native collagen. The result is collagen that is fully structured, it's consistent batch to batch. We control all the variables in the process and it carries the highest safety profile. And the result is 100% human-identical collagen. And we tested it on cells, and the cells noticed the difference. We tested it on our collagen on human lung and bryonic fiberlasts, and they propagated much faster on our true human collagen than on bovine collagen, as you can see here in the chart that uh we illustrate. Same cells, same conditions, the only variable was the collagen, and we get much faster, much better proliferation. And it also showed up in more complex systems, too. We tested it on human skin organelles that were grown in two metric composition compositions. Position, one without true human collagen and another one with red tail collagen. After an induced aging acceleration, the organoids grown on red tail collagen showed twice the aging mark. Same organoids again, same inducer, only different collagen source. Clearly, the ECM makes a difference. And the final piece of the solution, the scalability. Now, technology is designed to be industrially scalable. A single 200-liter bioreactor could produce up to 60 kilograms per year of collagen and can be replicated as many times as needed to meet any demand. So, with this level of scalability, true human collagen finally becomes competitive to any animal-derived biomaterial, solving the biggest challenge for making human collagen the go-to biomaterial for medical application. So to summarize, we believe that the future of medical biomaterials are human biomaterials, and Arrachis Bio is unlocking human biomaterials at scale for bioprinting and other medical applications. If you're working on tissue models or regenerative products, I generally like to talk to you. My email is here on the screen, and that concludes my presentation for this panel.

Collagen Q&A On Regulation

SPEAKER_01 23:08

Thank you. I'm happy to take questions.

SPEAKER_00 23:10

Well, thank you, Fabian. So it looks like within the first 30 minutes of this session, we had Scott speaking about polymeric 3D printing. I opened up about HIMED and our bioceramic 3D printing. And Fabian just gave us a very interesting talk about collagen. Quite a few convert quite a few questions in the chat. I'm going to begin with a question that I have. Do you see a bigger opportunity, Fabian, with 3D printing collagen directly or using it as a biologically component that can activate, say, 3D bioceramic or 3D polymeric-based surface?

SPEAKER_01 23:47

I would say you have to start with the problem that you're trying to solve, right? There's no one solution that fits all. And I definitely see this being the right answer, depending on the target application. So if it if it's a soft tissue, for example, I could definitely see collagen being the superior solution. But you know, having to have potentially significant load bearing, although nature shows us that you can do it with collagen, basically, our bones, our collagen with some calcium, some minerals, but we still don't know how to get that level of load bearing. So a combination of a hard potentially material with a coating of human collagen to enhance the regeneration, to improve the integration, I think that would be the ideal solution.

SPEAKER_00 24:38

I see. We have a question from Liv Herndon inquiring about regulatory pathway to success for this type of technology. Maybe you can spend a few moments speaking about the regulatory pathway.

SPEAKER_01 24:51

So we're we're early and we haven't basically cracked that one. But to begin with, uh regulators, as I mentioned, are in favor of replacing animal-derived materials with with you know human relevant. So I think the you know the time is in our favor. But since they you know everything that's new to the regulator is always suspicious, it it might take some some convincing to do and and to find the real the real path. Because when we talk about when you talk about collagen today, you automatically are being categorized as an animal derived material and not as a biologic. And and we're neither one or the other one. So it probably will take some some pathfinding together with with the regulators. But I think the tide is definitely in our favor. This is where the regulators want the industry to follow.

SPEAKER_00 25:41

I would agree. People probably are looking for you to lead that way. I I see a few interesting questions earlier. Jonathan Polak began the QA. He asked, is this the same team that did Cardi Heel and Agile C?

SPEAKER_01 25:56

Different team. No, no.

SPEAKER_00 25:58

Different team.

SPEAKER_01 25:59

It's not a nice piece of paper, no, no plastics.

SPEAKER_00 26:04

Okay. Okay, and then another uh straightforward question from Nita uh Latifi. How many days does it take for your treated cells to reach a 50-time collagen level in 2D culture?

SPEAKER_01 26:16

So we actually have a 3D biomanufacturing process, not a 2D biomanuf and it and it takes the cells to get to and I would say a couple of weeks to get to that full full potential. But we have a continuous manufacturing process. So those two initial weeks actually disappear and when when you look at the industrial continuous manufacturing process.

SPEAKER_00 26:40

Okay, very good. Fabian, please share your contact information in the chat so that folks can reach out to you if they so desire afterwards. And just a note for everyone in the audience: if you have specific questions, it's easier to send those questions forward in the QA box. It makes it easier for us to track, and this way we can indicate precisely the timestamp as to when those questions are being responded to for those that listen to the recording thereafter. Uh but Fabian, thank you very much for taking the time. I know it's late afternoon on your side in Israel, but it was a very interesting conversation. So great way to kick off uh today's event.

SPEAKER_01 27:18

Thank you, thank you.

SPEAKER_00 27:20

All right. Next up, we have Dr. Kamal Chadhari. Professor Chadhari is the assistant professor at the Department of Material Science and Engineering at my alma mater at Johns Hopkins University. I had the pleasure personally of listening to a talk that uh Kamal had given earlier this year. We've we were reflecting this earlier in the conversation, and he was in Hawaii uh following a talk that he had given at the Material Research Society, MRS. And the advisory board that I participated in, we were all very impressed with opportunities for AI and machine learning and contributions that Kamal has introduced specific to the material science department at Johns Hopkins University. To further introduce Kamal, he's an elected fellow of the American Physical Society. His research focuses on atomistic material designs using classical mechanics, quantum mechanics, and machine learning methods to accelerate experimental discovery. Kamal earned his PhD in material science and engineering from the University of Florida in 2015, and he previously served as a staff scientist at NIST. He's a creator of the widely used Jarvis at Tom GPT and ChatGPT Material Explorer Infrastructures for Materials Designs with over 200,000 users worldwide. And I might have missed a zero. I think I am correct, but 200,000. I know we actually had quite a few questions in advance of this session, specifically related to computational materials. So I personally am very excited to hear your talk, and uh the floor is

AI Tools For Materials Discovery

SPEAKER_00 28:58

yours.

SPEAKER_05 28:58

So thank you, Craig and Jenny, for the invitation. Let me try to share my screen. All right. Can you confirm you can see my screen? Yes. Yeah, okay. So yeah, good morning, everyone. It's a pleasure to be here in this exciting online conference. So, in coming 10-15 minutes, I'm gonna give a brief overview of what we're trying to do in Atom GPT lab or Chaudha Research Group using AI and physics for material design. So, again, thanks for the introduction. And as in the introduction, I'm a professor in material science engineering, electrical and computer engineering, data science and AI Institute, Ralph O'Connor Sustainable Energy Institute, and also a research associate at NIST, National Institute of Standards and Technology. There's a group website. If you forget about this, feel free to check out this group website and this is my group project website, atomgp.org, and there's something I developed at NIST before joining Hopkins.rbdish.gov. So, what do we do here in our group for AI and physical material design? If you are an experimentalist and you perform X-ray diffraction, Raman spectra, microscopy experiment, it's very difficult to find out uh sometime what are we measuring or what are we characterizing. Why can't we have something like a chat GPT where for material scientists where we can predict is atomic structure, property, performance? It's very hard. And that is something a foundational challenge in material science, which we try to handle is using uh AI-based materials design. Another thing, if you are a computational person, if you try to model a system atomistically, again, computers do not know physics, chemistry, material, they only know numbers. So, how we can impart our sophisticated information of physics or materials to computers is a challenging task, and that we try to do using a graph neural framework that I'll briefly touch upon a line framework, a line FF framework, and something called tight binding framework or SlackONET, which is a differential tide binding framework for quantum phenomena. All these models that I'm gonna briefly touch upon is available on a website. It's like a chat GPT for materials called atomgpd.org. It is has about 100 plus domain-specific apps, about 10,000 uh plus users. Jarvis has about 200,000 users, and with 50 plus institutions participating in this. So, without further details, let me go in a brief overview of what these things are and how they are being useful and being used over the community. So, first let me discuss why building a chat GPT of material site is so hard. First of all, if you look into the periodic table itself, there are 100 plus elements, and even if you look at the combination of this at the at the level of 10, the number of combinatorial possible materials is nearly infinite. So there is no complete data set, so there is no complete AI for this. Number two, in in usually you are dealing with heterogeneous data set, whether you are looking at angstrom level or micron level or very small time step or very fast time step, you'll be dealing with different types of modeling tasks. Experimentally, you might be dealing with like very small scale phenomena at nanoscale or micro scale or bulk scale and and the different quantity. So you have very heterogeneous data set, and such data set does not exist in completeness for material science, so there's no AI model which can be complete in that sense. Also, if you look into this paper, got 1500 scientists leave the lead on reproducibility. Most of our materials papers are in scientific papers are non-reproducible. More than 70% papers are non-reproducible. There are several inputs and output formats from experiments and simulation which are non-compatible. And some of these experiments can be very time consuming, very resource consuming. So these are some of the reasons why building a chat GPT for material science is so hard. This is a clip from my favorite movie Iron Man, where Iron Man is trying to go over the periodic table and try to build an art reactor using periodic table elements. So he probably selects something like cerium, uh diaspora, and does something called processing structured property performance. This is the key mantra or key guiding principle of material science. Even at the scale of movie or sci-fi movies, it's so difficult. Think how difficult it would be at the level of experiments or reality. This is a reason why US government and many other governments around the world have invested billions of dollars to actually this task. One of the first ones is Material Genome Initiative, $400 million initiative, US Chips Act, $52 billion initiative, and now $5 million initiative from Gen SS Mason. One of this initiative that has come out from this is Jarvis or Atom GPT infrastructure. Serves about 200,000 users as mentioned. About 100,000 materials. We have about 3 million data download and code download, and this is the highest fixed share download in in terms of material science and about 8,000 citations. Now, one of those infrastructure as available here at atomgp.org. You can log in with your email or gmail, and these are the peer-reviewed articles if you want to learn more about this. What does it provide? It provides a natural language model or vision language model, but also most importantly, it provides 100 plus domain-specific apps that you can use from searching materials, predicting properties, doing extra diffraction, and so on, and building custom-made tools. This particular tool was launched last year and about 10,000 users. And there are several YouTube tutorials about this that you can get familiar with. And as I mentioned, every day we develop a 100-day 100 app challenge for material science, starting from materials visualizers to extradifraction to Raman to STEM and so on, all free of cost and publicly available. So once we develop this, we can integrate this in ChatGPT itself. So when you asked ChatGPT Material Explorer, which is a paper we published last year, when you ask a question, instead of talking to ChatGPT model, it can go to custom model like a Jarvis or Atom GPT and give you a response and also explain what those things are. You can also connect Atom GPT with Cloud. So if you go to Cloud and go to settings, well, you need to have an account in Atom GPT first. Suppose you have an account in Atom GPT, and then you go to Cloud and connect using something called connectors, settings, connectors, yes, connectors, and connect using custom connector atomgpt.org. What is called atomgp.org slash mcp. I believe slash mcp. So mcp stand for model context protocol. This is the language how computers talk to each other in AI language. And then you have when you ask a question about material science in cloud, now it knows hundred plus domain-specific tools of material science. Suppose I'll ask a very simple question, fine aluminum off site material. So is it will talk to Atom GPT first and then give you the response instead of giving its own response, which can be hallucinated for a material design task specifically. So you can see it's it's pulling the data, making a summary, and then providing the result. Not just searching materials, but it can do several other tedious tasks. So this made a lot of news at Hopkins. So if you are familiar with Johns Hopkins Hub, they published an article about this AI system can be useful for material design staff. Now, if you are new to this field and you want to learn more about AI landscape and materials, we wrote a review article called Recent Advance in Application of Deep Learning Methods and Material Science in Nature. And it was published about four years ago. In just four years, you got about 1200 citations. It talks about chemical formula, atomic structure, text, spectroscopy, and uh and microscopy and how to use AI for this. One of these techniques that we use is called graph. So again, computers do not know physics, chemistry, math, materials, they only know numbers. So if you want to represent an atom atomic structure to computer, you can represent using graph. So if you look at this cute little picture here, which is silicon blue atom and oxygen red atom, you can distinguish silicon and oxygen based on their atomic number, boiling point, heat or vaporization of silicon versus oxygen. And this is one of the ways to distinguish through the computer this is silicon and oxygen. Now you also have to represent something called atomic bonds, and the atomic bonds can be represented by a bond length. Now different materials will have different number of uh atomic bonds, so you can use something called radial basis function to have a concrete representation of entire bond structure. This was fine in 2018, but then they did not have a very important feature called bond angles, and this is what we try to capture using this alchemistic line graph new network, and then we found that this can improve the AI model prediction up to 44%. This paper has been cited about 2000 times, and we use this model to discover a competition discover new superconductors and also a metal organic framework for carbon capture that we later synthesize and actually validate it. So now this model is available in Atom GPD or align. Now I show you a demo. Suppose you give a material like menosemboride with this lattice parameter, this coordinates, it can predict you properties like bulk modulus, formission energy, and stuff in a few seconds instead of running experiment for several months or days. You can use it for atomic structure optimization, you can use for molecular dynamics simulations such as what will happen at zero Kelvin, but also at room temperature, at high temperature, high pressure, and this kind of scenarios. You can also use this for something like band structure prediction for dynamical stability. So this is atomistic line graph neural network where we try to simulate atomistic systems using graph, but later we want to also predict electronic properties. So this is where not just atomic bonds and angles but also electronic orbitals comes into play. And we use something called slackonets, later poster tight band neural network that can be used. Now, if we provide uh atomic structure using lattice parameters and atomic coordinates, it will predict you like band structure, whether it be insulator, whether it be a metal, whether it will be a superconductor, whether it'll be a heat sink, and so on, this kind of property just based on a few clicks. Now, these are from theory, how about experiment? And this is where we develop a framework called diffract GPT, where you if you have an extra diffraction pattern, suppose two theta intensity, and you do not have a foundational model for this, uh, you have a lot several fitting procedures. Instead, you can use diffract GPT framework where you describe diffraction pattern in terms of language. This is a chemical formula, this is a two-theta intensity. Can we predict the output which is a atomic structure? So here is a demo where I put lap sticks lanthanum borides with two theta intensity. Of course, this material is a well-known material, so you can find in database. But now, suppose we do not we have a material we which do not exist in a data set. Okay, so now I'm gonna put a random material called we can also do something called writeable refinement and fit X diffraction pattern. But what happens if I come up with a new material called LAB6N, nitrogen? This material does not exist. Now, if you run this pattern matching, this will not be found because this material has never been discovered before. Because now we have diffract GPT model. I'm gonna select diffract GPT, it will generate atomic structure with reasonably high accuracy. You can see uh it generated a material with lanthanum, boron, and nitrogen. Now, this material model was published in this paper, and this model has been available on Hugging Face. It's a collection of uh AI models, and this is it has been downloaded about 240,000 times. This is more than deep seek download, if you know about deep seek AI model. This was one of the first times where AI from materials model become so popular that it can beat a mainstream model. Another important aspect in our group that we tackle is reproducibility. As I mentioned, most of the papers that are published nowadays are non-reproducible. So we developed a framework. This was funded from a project at uh NIST when I was a program leader there, and uh, this was about $5.2 million project. So we developed a framework to enhance reproducibility, transparency, and for various domains from AI, electronic structure, quantum computation experiment. Uh, you can learn more about this in this Jarvis E. Robert platform, how to enhance reproducibility. So uh I would finally would like to acknowledge team members, my grad students, and also undergrad. If you want to know more about my group, this is my website. I'd also like to acknowledge funding from NIST. Uh, we recently got a four uh part of the $400 million award from National Science Foundation program Cloud Lab and also the Eogenesis Lab. So, with that, uh feel free to check out my website, atondp.org. And uh, I also have several YouTube videos about demos on my YouTube channel, but Dr. Kamal Chaudri. Feel free to connect on LinkedIn, I'll subsfortify channel. Uh, this is my GitHub badge page. And also again, thank you for your time and attention, Craig and Jenny, for presenting this talk and happy to answer any question if you have.

SPEAKER_00 41:34

Thank

AI Limits Data And Loop Closure

SPEAKER_00 41:35

you very much, Dr. Chadhori. That's certainly a lot of information to consider and really amazing to see how AI and machine learning can contribute to these types of conversations. For those folks in our audience that have any questions, again, as a reminder, feel free to type them into the chat and we'll be happy to select them. A question that I have for you just what do you see as the most limiting factor right now, whether it be the AI models themselves or the lack of, as you were describing, the experimental data sets that would allow for that data to be leveraged on for these types of conversations? So, what I mean by that is that, you know, when we talk about materials properties, we're talking chemical formula, atomic structure, melting temperature, versus, say, you know, conversations using data like that for inorganics, the biocompatibility, the degradation rate, the bioactivity. Where do you see that transition? And do you see that more of a limiting factor? Or maybe the first uh concentration needs to be making these AI models as robust as necessary?

SPEAKER_05 42:42

Great question. So both aspects, so data. So we have data from materials, but they are very nuclear about a particular domain. So they have a lot of data about pharmaceutical energy, bulk modular, but not too much data on biocompatibility, for instance. So you have too much AI for materials going on on a certain area, but not on a larger area. That's one problem. So data is definitely a problem. And second, is closing the loop. So uh there are a lot of AI models which are trained on simulation data, but unfortunately, we do not have too much experimental data which are diverse enough. And so we cannot close the loop uh for theory and uh experiment AI AI driven driven loop. And this is another challenge because we have certain characterization uh instruments, but they are much slower than the speed at which simulation goes. So, for instance, X-ray diffraction, as I was mentioning, is one of the pre-techniques for material science. And unfortunately, it's very tedious even now to accelerate this using uh AI models, especially for defect system, for very generalizable system. What we presented here was for perfect system, for few defect systems, but it's not very generalized for a vast variety like biomaterials and so on. So we need to extend those AI systems from multiple domains. See their strengths and challenges and have a feedback loop with experiment. Unless we have done this successfully for various domains, it's very hard to have a very successful AI for material challenge, right? So data and also closing the loop. Those are the key challenges, I think, in my opinion, which are lacking right now.

SPEAKER_00 44:17

I have a second question. So I'm sharing with everyone in the chat a link for Johns Hopkins' Artificial Intelligence for Materials Design Laboratory. And I've had the pleasure of seeing this laboratory a few different times. This is something that was newly established at the university for being able to, as the website indicates, a closed loop facility that combines AI, machine learning, and data handling with automated automation and being able to characterize materials, characteristics such as indentation and phase composition and so on and so forth. I'm sure that there's a lot of folks that are in the crowd that would be enthusiastic about learning a little bit about this facility. So maybe you could spend some time talking about what this resource has to offer.

SPEAKER_05 45:05

Yeah, so I mean MD Lab and several other labs, MCP and so on, which are trying to accelerate autonomous experiments, especially for materials under extreme conditions. This is recently, it's been we are very fortunate to get a award from National Science Foundation, which is uh programmable programmable cloud lab, $400 million initiative, as I mentioned. And it it's it's trying to connect the same thing, like you know, uh connecting theory and experiment to discover or design new materials. And I think it's very hard to give an overview in just a few seconds or few minutes. So uh it's good that Craig is giving the links. Feel free to check out more and the links and reach out if you have any questions about those.

SPEAKER_00 45:47

I think the YouTube video does a tremendous job, but you're right. It was a it was a tough question to answer in a limited amount of time. We have a few two more questions that I see that popped up, which are also very applicable for the conversation. Um, Mazir asks the question of great talk, Kamal. How do you account for measurement and characterization errors in experiments when developing these models? Also, what is the typical accuracy of Adam GPT's final predictions?

SPEAKER_05 46:16

Yeah, so unfortunately, I didn't have time to go through the accuracy of this, but you can find, for instance, the frag GPT performance in the papers that I linked to. And it showed that it can so that we measured the accuracy of the model based on a couple of parameters like lattice parameters and the fractional coordinates. So we suppose we have 100,000 materials that we are training the frag GPT model. We do like a 90% training, 10% testing, the data that is never seen, and then you try to measure how accurately it predicted the lattice parameter, how accurately predicted the bond length. And we found that it can be pretty accurate, especially if you give chemical formula into account. If you do not give chemical formula, only give chemical elements or no chemical elements, then it's not very accurate. So uh the more priors you give, you get accuracy on the level level of 10 to 10% or 15% accuracy, which of course depends on what property or what model you are training. So it's not one thing fit all. But at least for that model, we can get up to 10% accuracy, which is a good start compared to having nothing. Uh same for a computational model, like we have very good model for creating formation energy, total energy, but not so good model for creating electronic band gas, for instance, right? So these are the challenges why we need to include more physics or more material science domain knowledge to have a substitute of having less amount of data. So when you have less amount of data, you put more physics, get more improvement. So these are some of the metrics again you can find in the papers in Diffract GPT, in Align, SlackONET to get more accuracy measures. But we have very concrete baseline and definitely is performing two to five percent, two to five times better than the typical baseline models, which is uh kind of take-on message here.

SPEAKER_00 48:02

Yeah. Okay, thank you. And the very last question from Liv Herndon with AI making materials discovery faster, where do you see the biggest impact for companies actually trying to develop new materials into production?

SPEAKER_05 48:16

Yeah, I think uh that's a great question. Uh there are several facets you can focus as a company. So, for instance, you can uh if you are a a materials company work also working in AI, you might sell your foundational model for like XFraction or foundational model for microscopy or uh foundational model for making real new recipes of that kind of class of material. That's one, or you can sell custom data set that your company specializes in, like biomaterials, or we heard about collagen, so collagen GPT, I'm just making it up. So there are a humongous opportunity if what you are working on and you can have a foundational model specialized for that particular domain, you will have a huge uh opportunity in in whatever field you are working on. So if you are a company, I think I would think of making a custom-made model or an API or a model context protocol for that particular niche, and you will have uh plenty of success, in my opinion, in the future.

SPEAKER_00 49:16

Well, very good. It was a pleasure to listen to you present. I think our audience found it very interesting based on the variety of questions that we received. So thank you very much for your time.

SPEAKER_05 49:25

Thank you again.

SPEAKER_00 49:26

Okay. Continuing right along, we have two more presenters lined up for this session. Next, I'd be happy to introduce Andrew Weems. So, Andrew, feel free to prepare your PowerPoint. And while you're doing so, I'll introduce you. Andrew is a resilient founder and CEO and a former assistant professor from Ohio University, specializing in medical devices, degradable polymers, photopolymers, and biomedical engineering. He's a senior member of the National Academy of Inventors, and he previously founded the UK company 4D Medicine, developing photopolymer resins. He's worked on Class I, Class II, and Class III medical devices. At Resilient Medical, Andrew has multiple patents related to tissue scaffolding devices, and he has led the commercialization efforts of the Bravo device platform, including regulatory strategy development, fundraising, and product development. So, Andrew, you have a very interesting background and an interesting topic to present to us today. Um, we look forward to hearing your discussion.

Lumpectomy Scaffolds And Go To Market

SPEAKER_04 50:31

Thank you. And thank you all for having me here today. What I'm going to talk about is a combination of what Resilient is actively doing and some of how we approach the problem of the need in breast cancer care. So we'll first talk about the clinical issues, and then we'll talk about how you might approach this if you were tackling a clinical problem, and then I'll share a little bit of what we've done and where we're at now. So one in two women are going to have an encounter with breast cancer in their lifetime. This may be a lump, this may be actually being diagnosed, but that that ultimately translates out to in the U.S., one in eight women having breast cancer at some point during their lives. In 2024, more than 370,000 women were diagnosed in the U.S., and that's expected to grow to nearly 450,000 new cases by 2030. Interestingly enough, though, there were more than 500,000 lympectomies that were performed in the same year. So the number of procedures that are being performed are outstripping the number of diagnoses. And there are a couple of reasons for this, including that patients that have been diagnosed previously are getting surgery, but also problems with both the diagnosis in terms of the surgery that's performed and with the procedure itself. And that's that's interesting because of nearly half a million surgeries, that's almost 3% of all the major US surgeries that are performed. This is a major, major surgery that people understand. You know someone who's had a lumpectomy, and yet we still have the same problems with it that we were having 20 years ago. That includes cosmetic changes to the breast site, pain associated with healing, risks of secondary cancers, issues with targeting, a whole litany of different morbidities are associated with this breast cancer surgery. And so we tried to come up with a solution to this. And so what I'll what I'll start with you is what would you consider to be clinically important for tackling any sort of problem? But in this case, what are the key design features that you need to approach a device going into the breast space? What are nice to have features, what are critical design features, and then what have other people tried before that you can build off of? And this does tie back to the materials in a moment, if you let me kind of meander around. So if you're looking at what's being done in today in the surgical suite, there are a host of different devices that are available. This is this is based on my evaluation of benefits. So this is not official, this is not published, this is just my opinion. But there's not a lot that's taken into account the cosmetic outcomes and the resorption time and balancing that for dealing with breast tissues specifically. The products that are on this map are very effective in certain indications. They've performed very well. A lot of them are still currently available and have been so for years. So what I'm not saying that there are problems with them overall, I'm saying that they're not designed specifically to address lumpectomy. And so if we want to come up with something that's going to work in dealing with a lumpectomy procedure, we want to have something that's going to take into account patient healing, cosmetic support, radio targeting. A lot of the things that have been done today don't do that. But what could we do? Well, we could look at foams, injectables, meshes, or, as we're here today, 3D printed scaffolds. Those are all very readily available and could be ways that get us to where we want to go. But you want to consider more than just what's going to be effective for addressing the problem. You also want to figure out how are you going to commercialize this, what's going to get you across the finish line in terms of the regulatory aspect and what's going to be able to be sold. So if you look just at the regulatory side of things, you want things that regulatory agencies are going to be able to understand and to comprehend. You also want things that are going to be relatively low risk, however, you're choosing to define that. And this is this is a very high overview, so there are always going to be exceptions to the rules. But generally speaking, that's staying away from personalization, that's staying away from drugs or active compounds, that's looking at well-understood, well-characterized materials and leaning into for US focused products, FDA's guidance on additive manufacturing, biocompatibility, risk assessment, and seeing what other people have done in the field as well. And again, that's just at a general level, but that gives you a good place to start. And for example, polyurethane foams. I'm my background was in foaming as a grad student. And so if you're talking about a void-filling application, my first thought generally is foams. And polyurethane foams are used today in different void occlusion applications. But in the breast space, they are associated with a fairly well-known failure. The shell around permanent breast implants was associated with certain cancer formations, and these were pulled from the market a couple of decades ago. That is still something that FDA is going to look at when it's being brought new devices into this space, especially if it includes a polyurethane component. And so you want to think if you're designing for a lympectomy device, how do you minimize some of your difficulty? I'd start by staying away from polyurethane phones, for instance. But we looked at other phones, you know, thiopoxy foams. This was something that I dealt with as a faculty member. We came up with a printable way to have a reactive foam. They were very biocompatible. We tested them in animals over a couple of months. We characterized how to develop the foams, pore sizes, controlling properties, tunability, things like that. But in talking to FDA, they were concerned that this was a new biomaterial. You know, this was something that they hadn't seen before. And so, not that it was not going to be regulated, but it was something that they were going to be very interested in understanding the nuances of its performance. And so the path to market was going to be more complicated with a brand new first-in-class biomaterial. And that's the big takeaway here on a novel material is not that you can't get it regulated, because you can, but you will have to go the extra mile at least to show that you understand the risks, that it can be well understood, it's reproducible, and that ultimately it's safe for the patient. And it's going to be up to you to show how that is actually the case, but that's that's going to be what your company is having to undertake at that point. The other thing that we did as in my group was we looked at designing new degradable polymer systems compatible with 3D printing. And so polyesters were where we started. We were making a whole variety of new materials using ring opening copolymerization so we could achieve different functionalities. In this case, we were looking at the alkene groups and making sure that they were reactive and so we could cross-link them in a DLP digital-like processing process to get these porous scaffolds. We had a lot of control. We had a lot of scalability available to this because we were doing it in a flow system, so it was a continuous production process, but new material. And even more so, to get to the mechanical properties that we want, we had to add in a diluent, and this is the diluent that's shown here. So there were a couple of steps in this process now. It's becoming a lot more complex. Again, we're getting really good biocompatibility. Shown here are some 3D images. We had cell proliferation in 3D, we had mechanical properties with an elastic modulus with less than one megapethcl. Really good in terms of looking at breast tissue, but new material, biological activity from the salicylic acid moity potentially, it was going to be tough to regulate. It was going to be problematic. So ultimately, we took a step back and we decided that we wanted tissue ingrowth, we wanted controlled degradation, and we wanted to focus on the healing aspect of the Bravo. So we're working with commercially available off-the-shelf polyesters that are processed through conventional 3D printing methods. So FDM type processing because it's cheap, it's scalable, and it's well understood. It also lines up with FDA guidance and is something that we can show pretty readily how it's going to be safe. Also, we have designed the product to fit in with what's being done in the clinic. So it fits with clinical workflow, it's intended to reduce surgical time, therefore patient risk, therefore procedure costs, and it's not adding any new steps. And that's the other aspect of commercialization is is somebody going to buy it? If it's new, if it's exciting, that's great. But if it's going to cause you to redesign the whole surgery, you're going to have a hard time making the sale. If it fits with what people are doing now, if they're not having to innovate the process, they're just having better outcomes from your new technology, they are more likely to take it on. So we did uh a slew of testing required by ISO 10993. That is standard guidance for biocompatibility. Ultimately, what we found is that our device is non-pyrogenic, non-irritating, non-sensitizing, biocompatible, all these things are the requirements that you have to undertake in order to move towards the goal. We did a lot of studies in animals, and so this is where I would push back a little bit on some of the previous comments made on this panel. Animal testing, while there is interest in reducing the burden of it, is not likely to go anywhere for quite a while. Not necessarily because it aligns perfectly with human subjects, but because it's a well-understood set of tests. We are able to make comparisons with other products very readily that we can't make with bench top models yet. Animal testing just gives us that comparison. And so there are there are strong advantages to it, even if it's not a perfect analog for human testing. But you know, we did quite a bit of it. And so ultimately, what we were able to put together is a full life cycle analysis of our product so we could relate the clinical endpoints of our problem. So what are the things that they're doing in the surgical suite to treat the patient most effectively? And then how does our device performance tie into this? So we know pathology at 90 days, at 180 days, at 365 days, so a year, we know mechanical performance at these key points. We know when the patients are going to be imaged typically, we know when they're going to get radiation therapy, and we can tie all of that together because it's not just a question of how long does it take to resorb, what's the mechanical strength. It's it's all of the package. How do you make a product from start to finish, especially if it's resorbable, that is going to be safe and perform the way that you intend it to, that starts with your material. So again, really quick high-level view of what we're doing and how materials tie to it. I'm always happy to try to share more if I can, certainly talk about what it is that we're doing, or if there's opportunities to work with other people, definitely happy to explore that too. My email address is here and I can put it in the chat. Happy to answer any questions. Thank you for

Academia Versus Startup Mindset

SPEAKER_04 1:01:47

listening.

SPEAKER_00 1:01:47

Thank you, Andrew. Andrew, I tried to make it clear in my introduction, you have a very unique background, right? And I know that there's a lot of folks that are on this call that uh not just come from industry, but come from academia, and you have a great deal of experience with both, right? You were an assistant professor for a number of years at Ohio University, and then it looks like you left Ohio University to establish your own company, which it seems like is performing quite nicely over three to four years. Maybe you can talk a little bit just about that thought process and and uh you know where you see, I'm sure that you perform better in your job as a co-founder or founding member of uh Resolute Resilient Medical, just coming from the academic background. But it's always interesting to learn and hear directly from our speakers if you can talk a little bit about your your background. Sure. Who who doesn't like to talk about themselves?

SPEAKER_04 1:02:42

The the thought, so generally speaking, there is a very different mindset in my experience from what academics do versus what industry people will do. And it's not that one is necessarily better than the other, but one is tailored to the you know the field that you're in. In academia, we do a lot of concern, we are concerned a lot with how are we going to test for some particular facet. You know, we we've identified some problem. It may not even be a real problem, but academics have found the problem that they're gonna solve. And then they they do the experiment to target it. And there's a lot of optimization, there is a lot of testing, you're trying to make things better. There may not be a fixed endpoint. In commercialization and industry, perfect is the enemy of good enough. And you you want to get to the place where your product can do what you say it can do, it can solve the problem that you want to have, and then you want to get it out the door. So that doesn't necessarily mean that you're always doing testing, you're always gonna be refining. You're gonna get to the place where it performs as required, and then you want to move it forward, not only because that's how you're gonna make money, because it is, but you're also spending money. And I I don't have a university budget to fall on anymore, which means if if we're if we're cash out, it's not a good day. Whereas in academia, you know, there's a TA position if a student needs to be paid or something like that. So it's very different stakes that way. Um, the other thing that I'll say is that in academia, my experience was we are we are doing a disservice to the many students that are going into industry. We are preparing graduate students to go into academia, and it would be it would behoove students today, I think, to prepare themselves to go into industry if they think there is a chance of them going that route at all. Uh because it is a different mindset. There are different requirements, and and you can do a lot to prepare yourself very effectively. It just may not be what you're necessarily doing in your lab all the time. Does that answer your question or at least go in the right direction?

SPEAKER_00 1:04:46

No, absolutely. It's uh I always find it interesting to hear, you know, those uh thoughts that, you know, talented uh folks make as they're progressing their careers. And uh I your background just caught me as very intriguing. So I appreciate you providing some more information there.

SPEAKER_03 1:05:00

Oh, thank you. Yeah.

SPEAKER_00 1:05:01

We have more questions rolling in. So Liv is asking a question similar to what I was thinking about, also pertaining to scalability. Uh Liv's question is can you compare the automated versus manual processes for removing printed scaffolds? And how much of the process have you seen automated at this point?

SPEAKER_04 1:05:20

That is a really good question because that gives you the ability to further reduce cost and improve your throughput, as you say. I have not come across a whole lot of automation in this process. The approach that we were taking had relatively low manufacturing costs to begin with. If you're if you're doing off-the-shelf materials, if you're doing off-the-shelf printing, those costs do start to reduce on their own just through basic scale up. The big thing that we see is that there are there are a lot of other factors that build into the final cost of a device. So you have packaging, shipping, sterilization, and then all your As well as your continuous process validation, those are a lot of places where your cost really starts to stack up too. Um and those may be a little bit less prone to deviate with scale up. Regarding the automation, there's only one printer that's coming to mind that that really allows for automation and it requires a redesign, at least what I saw for our parts. And so we didn't pursue it heavily as a result. There may be applications where it's absolutely perfect. It just wasn't a great fit for what we were doing. Very good. Thank you.

SPEAKER_00 1:06:31

I see some more questions. Some of these look very applicable for the panel discussion that we're going to be having at the very end of the discussion. So, Andrew, we're going to thank you for your time and we're going to move forward to our last speaker. But again, for those of you that are in the audience, if your question has not been answered just yet, know that we still have it flagged, but we're holding on to that question for the very end as part of a greater conversation, which will conclude our session. So next up is Ibrahim Yarali. So Ibrahim, feel free to pull up your PowerPoint and and get your screen set up. And while you're doing so, you can turn on your camera and I'll be happy to introduce you. Yeah. Ibrahim, where are you based out of?

SPEAKER_02 1:07:11

Okay. I Andrew, he may want to share as well.

SPEAKER_06 1:07:15

I guess should be fine now right. Maybe double click if you can get it.

SPEAKER_02 1:07:20

Yeah. I think I think you're I think it's just uh Everen's screen now, but somehow his presentation is not showing up on our screen for some reason.

SPEAKER_06 1:07:29

Yeah, it seems like something is wrong with my Zoom today, then it doesn't work.

SPEAKER_02 1:07:33

Try again.

SPEAKER_06 1:07:34

Yeah. Is it okay?

SPEAKER_02 1:07:36

Try again. Try to share again. Yeah.

SPEAKER_06 1:07:38

Okay, do you see my screen now?

SPEAKER_02 1:07:39

I I do see that you're entering this mode where you're about to share, but for some reason your slide is not showing up. Were you given an option of which screen to share just now?

SPEAKER_06 1:07:51

I the PowerPoint will probably maybe I need to like you share the full screen.

SPEAKER_02 1:07:55

Okay. When you enter sharing screen, did did did Zoom ask you to share if you want to do the the screen with PowerPoint or some other part of your screen? Are you using multiple screens at the moment? Okay, I think his internet is also kind of not working. Hello? Let me see.

SPEAKER_00 1:08:13

Yeah, it looks like his camera froze.

SPEAKER_02 1:08:15

Yeah. Okay. Okay, you're back.

SPEAKER_00 1:08:17

Yeah, I'm serious.

SPEAKER_02 1:08:18

Yeah. Yeah. I think it I think it was your internet issue. So why don't you try again now? You're back.

SPEAKER_06 1:08:24

Yeah, let me share my screen.

SPEAKER_02 1:08:26

Okay. Fingers crossed.

SPEAKER_06 1:08:28

Yeah. Yeah.

SPEAKER_02 1:08:28

Okay, Liv, um, would you mind of typing your comments in the last session if you have a suggestion? Try closing your camera, someone said.

SPEAKER_00 1:08:38

Or if you would like to email me your presentation and I could share it on my screen and page when you instruct me.

SPEAKER_02 1:08:45

Yeah, that would be uh I see. Bandwiz issues. Hmm, interesting. Okay. Yeah, Ibrahim, if you want to do that as a backup. The other option is turn off your camera and see if you can share your screen.

SPEAKER_06 1:08:57

Yeah, let me try it again because as soon as I share it and kicked out from the try it again.

SPEAKER_02 1:09:02

Okay. Yeah, no, we see it. We see it now.

SPEAKER_06 1:09:05

Okay, so probably the problem is the camera.

SPEAKER_02 1:09:07

I think it's a bandwidth, like the audience pointed out. That's the problem. Okay. Sounds good.

SPEAKER_00 1:09:12

Go ahead. So sorry for the grease that we had just Before you begin. I'd like to give you the proper introduction. But where are you out of? Where are you the name of the Netherlands? Yeah. Okay, wonderful. So we've covered all the entire world now. We've gone back to all, yeah. So Ibrahim Urali is a postdoctoral researcher at the Institute for Technology Inspired Regenerative Medicine at I'm sorry, I'm gonna mispronounce the name. A Maastricht University? Yeah, Maastricht. Okay, wonderful. His multidisciplinary research focuses on developing acoustic responsive 3D architected biomaterials that promote skeletal regeneration by guiding both uh chondrogenic and osteogenic differentiation of stem cells. I know that you have your PhD from the Department of Biomechanical Engineering and Precision and Microsystems Engineering at Delft University, and we look forward to hearing from you. So uh now that your audio and your your screen is working, uh why delay the inevitable, the floor is yours.

Geometry Driven Bone Regeneration Scaffolds

SPEAKER_06 1:10:13

Yeah, thank you very much for your uh introduction, and I'm very happy to be here today. Yeah, so uh I'm Ebrahim, as you mentioned, I'm from Maryland, uh yeah, from the Netherlands, and I'm doing my postdoc uh research here. Uh yeah, today I would like to talk about like architected biomaterials and uh like basically the role of geometry in the bone regeneration field. Before going to the like the details, I would like to, as I'm living also in the Netherlands, I would like to also share some numbers with you. Uh I don't know if you knew this, but in the Netherlands we have more bikes than people. So like 23 million bikes than uh 18 million that we have the population here. But of course, we have also a lot of accidents. So it's like 81,000 uh accidents that go into the emergency rooms from the bike per year, and usually like 74-72% of these uh bike accidents going to have uh bone fractures, and uh well we also have quite uh like a significant number of uh population for the like a people above 65, and uh that they also go through bone fracture uh like easier and also osteo prices, and uh yeah, because of it, like uh the other reason could be the tumors that like last in 2023, if I'm not mistaken, is like two 700 cases of uh sarcoma, and uh yeah, indeed other reasons like a genetic and lifestyle. But to to cope with this, uh we usually have two pro two solutions. The one depending on the the size of the bone tra damage, we use the implants, and there was also like 700,000 bone implants have already been registered in the Netherlands, and uh the second one is uh bone regeneration. So, what if we just just use uh scaffolds and biomaterials that they can regenerate the bone themselves, or also the other field, which is a bit far from what I present in today, is the cell therapies. But if we look at the in detail, if we look at also the the structure of the trabicular bone, it's quite porous and uh very random pore distribution, as you can see here, and they offer quite very different uh properties from mechanical properties to uh morphometric properties, mass transport, and acoustic properties. Well, yeah, so we were looking for like a material or like artificial materials or synthesized materials that could mimic these properties and at least give us more freedom to control to have control over these properties. So, a very good candidate for this one we call them meta-biomaterials or metamaterials. They are architected materials that the overall property coming from the how we like like it designed the micro architectures, not the base materials. So, for instance, and they can be also designed in different ways. We can design them in a ordered way, or we can design them in the stochastic or like a random way. So, for instance, in the ordered way, so it's like a repetition of a single unit cells that can be repeated in different directions. And for the random one, it's also could be you have a space in unit cells that you can refill it with random aspect or the object, and then you can make your scaffolds based on the properties that you want to have. But what are the problems like so far that in the this filling both of tissue and bone replacement implants and also for the the other aspect for the bone implant design? So, what we have uh in the during my PhD and the previous group, Amirza Purz group, into Udelph, the problem was uh yeah, like the conventional bone hip implants that they usually have is uh like when you talk to the surgeons, like after a few months, uh when you implant the bone into the implant into the body, they need to do resurgery again in some cases because the bone, because there is a gap happens between the bone at the interface of the bone and the car uh the the uh implant. So to solve this problem, we they noticed that is like the mechanically the bone could be designed uh better. So, what they did, so basically they designed the bone, the implant into different uh sides, and each size they experience different uh displacement field. And with this, you wouldn't have the like a the loosening problem at the interface between the bone and the implant. So, okay, you can change one single you can change the design in a way that you solve this problem, but the other the main problem is that when you change the geometry of the implant, then the other properties like stiffness, like other properties, also changes. So it compromises it. So that that's that's one issue. But then the other one is like in mechanobiological studies, in more fundamental in vitro studies, that it's also my kind of one of my fields, like the my field also that I'm working on is we are more interested to see okay, what is the effect of, for instance, a single like the isolated effect of porosity. If we have different biomaterials with different porosity, what are the effects? But the problem is that we need to find, for instance, uh two scaffolds that only porosity is the different, all other properties is the same. So the idea was how to break this like a property dependence in scaffold in general, in bone scaffold design. So, or more specifically, if we have, for instance, here a few different geometries for the bone as bone scaffold or micro environment, which one is better for the bone regeneration? So that was all the questions. So to answer these questions first, like how to decouple them, and then the second, what are the single effects of biophysical cues on the bone regeneration? Well, I was using also my mechanical engineering background where I graduated in mechanical engineering. But there are two methods usually. The first one is uh I call it forward method, is uh kind of conventional method. So, what I did was also what I did during my PhD. So I created a very huge library of unit cells and designs, like 50,000 designs, possible, like all possible designs. And then I did completion modeling on them, and then I did optimization, and then after the optimization, we did the we could choose the decoupled properties, like a decouple uh a very general method that can be used for every single properties. And the second one is using the inverse method, which is machine learning based. So the user would choose. So the user chooses okay, I need these specific properties for the elastic modulus for the porosity and for this, and then we don't need to really have a big library of unit cells, so the machine learning would uh design the scaffolds for us. So this is something that I'm also working on during my uh this project in the my uh postdoc. So today I just gonna quickly go through the the first method that how can we design like a patient-specific uh scaffolds, but with decoupled properties with the yeah, decouple properties. So for these, yeah, there was uh I'm not sure how I can move a little bit. This maybe I can yeah, okay. Yeah, so the I chose the unit cells, and then with the unit cells you have some like a boundary condition, and with this I could find out the I could drive the mathematical relationship all here, and then we could generate like 44 45 more or less thousands unit cells. As I was looking for a pair with decoupled properties, so the total permutation was around one billion permutations. And here, for instance, like I'm just changing the one one input parameters, and you see how, for instance, we can generate different uh scaffolds and different designs, like as you can see here that it changes. But for the modeling of these these structures, as they are uh like it's really and yeah, like if we use conventional modeling for this, it would take like three hours at least in one direction, and then in total, like uh nine hours at least. So it was a big problem because we had a lot of designs, like 45 designs. So then, yeah, so we were looking for some way. So I then I developed a method that we could generate all the properties in like less than one minute for each design. So then it was very quick for us, and with this, we could uh quickly generate all the properties, and we did go through all the validation, mechanical validation, mass transfer validation, morphological, morphological validation. So, for instance, here I just like like a yeah, here is just the mechanical properties distribution. With this, we just wanted to know how much design space we have, how much we can play around with the parameters, and in the end, we also did uh optimization, and then we could choose, then in the end, we could choose designs that they are decoupled, so decouples in terms of for instance porosity or other parameters. And then to manufacture these scaffolds, well, they are so complicated in in terms of shape, so like sophisticated microarchitectures, so it wasn't kind of not easy. So we did two photon polymerization technique, which is basically like uh like well, it's a very uh I would say advanced 3D printing technique that we can light base, and also we can uh print from uh 100 nanometers up to millimeters uh structures uh from a droplet of uh photoresist. And then uh we also did print them at the macro scales where uh we used polyjet, so we could print these structures at the very kind of bigger scales, as as you can see here. We also here is just a video capturing the the compression test of the the samples, and then we also the next step was to see the osteogenic responses of these uh scaffolds, the the two PP printed ones. So for this, yeah, we used very like uh simple cells for now, like cell lines. This is the my PhD that uh yes, we see the scaffolds over time uh with the osteogenic uh growth factors, and then we noticed that they they respond quite differently uh to the scaffolds, and some scaffolds they cannot deform them, but some they deform them quite a lot, and then uh uh we could yeah in terms of mechanotransduction and also to see how cells how much force applies to the scaffold by the cells. It was quite interesting, and we also had some conclusion that how like the shape, the shape of the pores, it is very important how how uh the scaffolds is designed to have a particular properties cell response. We also did some uh protein-based analysis, uh like uh Ron X2 and uh ALP to see how like a cells are are they cells differentiated in the structures where it was the case, and also we could quantify and uh find the significant differences between the geometries. And here is also just uh some animations of how the cells, the cytoskeleton morphology of the cells like look like on these scaffolds, and yeah, we depending on the geometry, we have a different uh cytoskeleton, cytoskeleton morphology for the cells. But so far, I was just talking about like a static, but as we are going to use these ball biomaterials into the body, and they also they are experiencing completely different environment, like very dynamic in terms of uh likely flow and also in terms of the the load that they applied to them. So that's why I was also interested to also go through how can we make structures, print structures that upon exposure of stimulation, they deform or they f the function changes. So one was we developed the hydrogel that could uh be that could respond that respond to temperature, as you can see here, at the very small scale. So by changing the temperature, the hydrogel would deform these beams, or the idea was to also make some little scaffolds that that that the deformation that they deform upon increasing the temperature. And there was also something just uh we I explored it just by chance, honestly. It was uh like we printed these very like noodle shape scaffolds that they are quite well that they are quite soft and like they're made of uh plastiomers, and but when you put them back into the bot into the water, they can re like uh get back to their original shape. But uh that's something that we are still working on. But also again, for the in terms of dynamic cell mechanobiology, uh, during my postdoc in Berlin under uh Professor Lorenzo Moronisbu. So we are going, we are now working on acoustic metamaterials where we can control the vibration that is induced in the scaffolds via the ultrasound, and we want we are working on the human mesenchomoster muscle cells to control the osteogenic and also chondrogenic differentiation of the cells. So so that's the idea. So we have different geometry. So basically, by the changing the shape and the pore shape of the scaffolds, and with ultrasound, we going to have different vibration in the system. And yeah, here is just like how our setup looks like. So we have different uh frequencies, and the setup have has been designed in a way that we have like a space for different parameters, like frequencies, like intensities, and here is also yeah, just how the samples look like here, how the samples uh behave under ultrasound intensity. Here I just wanted to mention because I know like in the panel here it was more clinical or pre pre-clinical, close to clinical uh approval and applications. But yeah, so what we do here, and also as was discussed by Andrew and others and Fabian, yeah. I mean what I did here was more very fundamental studies, but I wish, well, yeah, in the future we can also go step forward to the clinical applications, and for this, indeed, there is a big gap from like it to the translation from the in vitro studies to the clinic, and there is a very slow process which could be from both sides, from the regulation and also from our scientists. And uh yeah, with this, uh I would like to thank all those people who yeah, I was working with and you too. Thank

Acoustic Responsive And 4D Biomaterials

SPEAKER_06 1:24:11

you.

SPEAKER_00 1:24:11

Thank you very much, Ibrahim. Yeah. So I think let's invite all of our panel members to turn on their cameras and come back on for a concluding discussion for today's session. Lots of interesting information. Certainly it was a very diverse conversation on a broader biomaterial scale. And again, for those in the audience, feel free to continue to submit your questions. Jenny, I know you had quite a few questions that you had received in advance of the session. Is there anyone in particular that you have on mind or or how do you how would you like to begin this final session?

SPEAKER_02 1:24:44

Well, actually, I I think we should uh probably address the question from the audience first, since they've been waiting around for a little bit. I think usually, you know, we we want people to talk about the future a little bit, but I think we can talk about that later.

SPEAKER_00 1:24:57

All right. So it looks like there were a few questions that came in from our audience that there was one question earlier in the session for Professor Shudhari related to the use of AI and machine learning. The question from Rajan Shudhari. My question is does machine learning directly determine whether a material is biocompatible or it learns the relationship between materials characterization and synthesis conditions and measured biological outcomes?

SPEAKER_05 1:25:23

So in general, as I said, a machine does not know material science biology. You have to train that based on the data that you have. So of the bad, I'm not aware of a direct model that can tell you directly one and zero whether it's biocompatible or not. So you'll have to do your homework where you collect the data, train the model, have the foundational model for this. Right.

SPEAKER_00 1:25:45

So training of the model is paramount.

SPEAKER_04 1:25:47

Can we add on to that just a little bit?

SPEAKER_00 1:25:49

Sure, by all means.

SPEAKER_04 1:25:50

Two other considerations for that question. Biocompatibility has a pretty broad and nuanced definition that links to what your application is. So it would be interesting to know from the AI computer modeling side how that might be approached very long term. But I think in the short term now, even some biomaterials that are considered biocompatible in certain applications wouldn't qualify in others, or they have adverse events that uh reduce their biocompatibility so it isn't binary, anyways. But it also really depends on your application. That saying something is a biocompatible material doesn't necessarily have a lot of meaning so much as the application in which it's being used is biocompatible or the material is biocompatible in that application.

SPEAKER_00 1:26:38

Right. Very much application specific. That kind of relates to the question that I had asked of Fabian as far as the hydrogels or excuse me, the collagen, and incorporating collagen as the predominant resin, if you will, or incorporating that within a plastic or polymeric or or even a ceramic-based uh print. So very much application specific for certain.

SPEAKER_02 1:26:59

I also want to add a quick, and everybody is that the panels can certainly ask each other questions. I'm pretty sure the speakers have a lot of questions for each other, and I think this is a good good format to start. Now is you can also ask your own questions to each other, and that will be a very interesting conversation. So so the speakers can ask each other questions, and the audience can also ask us questions and and not just us dictating, you know, what to be discussed. It's uh, you know, it's just like a little group chat right now. I just want to make sure people kind of relax and feel free to ask each other questions. Okay, I think no questions. That's all silence after I said that.

SPEAKER_00 1:27:37

Andrew, you had one slide and and you said that maybe your thoughts on clinical testing and animal testing might be different from in in the context of what you were presenting. I wanted to ask you a question to elaborate more on that, but we were shy on time. Is there any more anything more that you would like to comment on related to that?

SPEAKER_04 1:27:55

So related to the use of animal models in testing?

SPEAKER_00 1:27:58

Exactly.

SPEAKER_04 1:28:00

So I know that Fabian has a different opinion than I do, and this I think would be a good place to hear his view on it as well. My stance is that with the with a lot of the regulatory pathways that exist and with the the definitions plural that exist for biocompatibility, animal model testing still makes a lot of sense and will, I think, continue to make a lot of sense for a number of reasons, including being able to compare performance with existing devices. So being able to actually metricize what you're doing and have a quantifiable or semi-quantifiable comparison, being able to show that to a regulator, so demonstrating your safety in a way that other people can look at it and can understand it. And then ultimately reducing the burden that would, if if we removed animal testing, would I think shift towards clinical testing too, that has a number of problematic components too. There's a reason why we're using live subjects, but they're not people to make these comparisons. So I understand there are really good reasons for moving towards bench top models, 3D cell culture, et cetera, but I don't think that we are far enough into that field to really have a good grasp on moving away from it. I know granting agencies are interested in developing the landscape, but my my understanding from the regulatory side is that that's still a very long way off because just because of comparisons and the body of work that exists now.

SPEAKER_02 1:29:30

Yeah, I also want to act because we just hosted an event two weeks ago on this very subject, the new approach methodologies. And I think the event is still on demand for four more days if you want to go watch it for yourself. And I agree with Andrew. I think it's, I don't know how far it's gonna be. I don't think we can personally, I don't think we can 100% replace animals, but you know, this field is growing, and we literally have thousands of these NAMs, the new approach methodologies, but we don't have enough validations to prove they can be equivalent or superior. It also is application specific. Depends on what you're testing. And it's almost like a thousand blind men are touching an elephant and everybody has one piece of it, but they don't have the picture. So the industry is very new and is evolving, and we're seeing a lot of exciting data, but we're not at a pivotal point where the majority of the animal testing can be replaced by these models yet. I think the directional is, you know, we are gonna use more human, similar, these bench top models, but you know, where can we eventually replace a multi-organ living uh living entity? That is questionable, but we can certainly get more data now with these models. So it probably complementary, so you may use less animals in the future, but not I I don't think we can replace it completely. And also, there is a long history of animal testing because people died uh when there was no animal testing.

Panel Debate On Animal Testing

SPEAKER_02 1:31:08

So it actually is a life-saving step in developing any biopharma or medical device process. So it's it's in it's critical. Just want to add that. So yeah, so we have an article I shared, we also have on-demand on recording for four more days. If you want to go watch it.

SPEAKER_01 1:31:26

Maybe I'll I'll add my my two cents here. I I wasn't trying to advocate for dropping all animal models, especially in in the types of applications we're dealing with here, which are, as you mentioned, multi-organ, complex, large. Well, the the the mechanical and the structural requirements cannot be reproduced with anything that we have today on on the on LANs or anything like it. But on the micro level, on the cellular, you know, tissue level, making you know uh sure that the material is, as was discussed earlier, really biocompatible for the application. And I know even going further than biocompatibility is regenerative in the application, at least that's the ideal, right? Um so to striving to get closer to that will get us better results in the end. But we you know, we we're nowhere near dropping animals altogether, and I wasn't even advocating whether all I was trying to say is that at the cellular level, at the tissue level, the cells know the difference between different biomaterials, and and we need to take that into account. What when we try to extrapolate from an animal result to a human result?

SPEAKER_00 1:32:40

I realize Ibrahim, we we're quick to skip over to the to the broader conversation. I had a question specific for you, and I'm wondering, you had so many videos in your presentation, which were really interesting to see. Maybe that was one of the reasons that bandwidth initially had caused complications. Could you talk a little bit more about 4D printing and specifically what material properties and material characteristics are most critical over time for the control?

SPEAKER_06 1:33:07

Yeah, thank you. Uh yeah, for 4D printing, I mean, well, it's a kind of new terminology in the field, but it's basically 3D printing. And we can either 3D print and like a smart object or biomedical device or like an object that with externalized simulation we could control the function. So this functionality usually we people talk about the shape morphing. So, like for instance, that there was the idea of making stands that in size there when you print them, they're small. But then we did also some individual tests that when you expose them to the a little bit change in the temperature, but again, depends uh also on the target and the tissue. You could also use different applications in ultrasound, or you could also magnetic field. So in terms of uh yeah, again, I guess I get the same notification, that's very sad. But yeah, uh, yeah, so uh but in terms of materials, yeah, we are very limited because we can only use a very limited number of materials that they can have these shape-shifting properties. And uh the whole concept is the heterogeneity. So we need to make a heterogeneous material that they respond differently to the simulation. That's why then they bend or they we jinxed it with the the bandwidth here.

SPEAKER_00 1:34:20

It is again.

SPEAKER_02 1:34:21

Yeah, I think I think I think maybe your video if you're off cam, like it's probably better, unfortunately.

SPEAKER_06 1:34:26

Yeah, I guess I need to turn out the video, yeah. Yeah, yeah, so yeah, so uh yeah, as long as we can have the heterogeneity into the material, we could uh control the shape morphing in the stuff in the 4D printed objects. And uh with this, also that's what I'm also working on with the ultrasound. That's also from outside of the body, we could also do the so if we have uh damage like from the osteocone, like uh from the bone and also from the cartilage, like for instance OCD or like osteoart crisis cases that they have if we want to mimic both sides, the cartilage and also the bone, so we could also make it smaller biomaterials that can fit into the damage area, and also with the ultrasound, we could also expand it, expand and refill the damaged area, and also we can also impose different vibrational modes for each side because right now what we are doing now we notice that for making a cartilage, at least for differentiating and guiding the differentiation of HMSCs toward cartilage and also toward bone, they we need to use different ultrasound settings. So they use different vibes, they need different vibrational uh profile. So, yeah, that's what we are working on.

SPEAKER_00 1:35:39

Fabian, I'm curious if you would like to comment on that, just considering collagen. As you began, collagen is a very interesting material, and when you consider crosslinking and what that might be for a 4D printing, I'm curious if you have any thoughts on what Ibrahim just said.

SPEAKER_01 1:35:54

I'd love him to try printing with collagen and and having all that, all those experiments and added functionality be tested on collagen. Eventually, the tissue, if we're talking about cartilage, we're talking about bones, if you look at the the microstructure, that's collagen plus some additives being naturally manufactured. So if we can get closer to that or even faster to the end result, because the end result is to have the natural tissue replace right the implant. That's the end result. And if you if we could start with something that's already similar and and conducive and and only regenerative to the end result, maybe we'll get faster, better, better results. And maybe combat combining collagen with with some meta materials would lead some. No, I'm um, yeah, that's that's fascinating to me. So I would love to see that being tested. But how to how to say uh I see no theoretical reason why you couldn't model collagen to to meet that and and print it this way, but I the devil is in the details always, so I maybe I'll stop short there. Happy, happy to engage after the the call, of course. Ibrahim, you're you're welcome to reach out and we can chat about that.

SPEAKER_06 1:37:10

Yeah, sure, it would be nice.

SPEAKER_01 1:37:11

It would be nice because I know it's quite kind of new and uh yeah, uh still uh we're also working on it, but it would be very nice here to also extend the especially the the ultrasound added functionality because the the mechanical pressures the uh definitely guide the tissue regeneration. So using ultrasound to guide that is fascinating in my opinion, and I I applaud that that effort.

SPEAKER_06 1:37:36

Yeah, thank you very much. Yeah, yeah. So uh yeah,

AI Hype Real Impact And Wish List

SPEAKER_06 1:37:39

let's see. So we are still working on it, hopefully it will be out uh next year, and yeah.

SPEAKER_02 1:37:44

Yeah, and also I just read that ultrasound in itself as a technology is also evolving, and sometimes meta material, which was a discussion by Abraham, was it can also be instrumental in innovating ultrasound itself. So the technology kind of synchronized with one another, it's quite interesting.

SPEAKER_00 1:37:59

Yeah, thank you.

unknown 1:38:01

Yeah.

SPEAKER_02 1:38:02

Any other I have a question, Craig. More about futuristic. You know, now is like most people want to learn about AI and robotics, and uh, we we clearly got a taste today uh with Dr. uh Chowdery's talk. It was fascinating. I'm pretty sure everybody's gonna already subscribe to his YouTube channel and hopefully to learn more about um his chat, he's uh Adam GTP. But I'm curious of with everybody else here with you know a business or research, mostly in the analog world. How what what kind of hope that you think this this tailwind from the AI industry can help with material material science innovation? Do you think this is a hype, or do you really have a lot uh a lot of optimism in it? And you know, what what is what is on the top of your wish list? And I'm gonna start with Dr. Childry because he's clearly optimistic. So kind of want to know what you guys think.

SPEAKER_05 1:38:59

I'm bound to be optimistic, right? So here because I'm working on this field. So um, yeah, I can start with the numbers. So if you look into some of the venture capitalists that are investing in the areas of tech in Silicon Valley, one of the most important areas they are investing right now is AI for materials. The reason is suppose I'm just making an example. If you can make a superconductor at room temperature, which is very difficult, it's a multi-trillion dollar economy. You never have to charge a battery, you never have to do XYZ and so on. So a small amount of invention in and tossed materials can have such a huge impact. And not just semiconductor, like semiconductors, of course, is fueling the AI and in turn is like a like you know, better semiconductors, better AI and stuff, and also in bio as well. So I'm really hopeful that people from different backgrounds will come along and uh tackle this in an optimistic way. And there are hypers in the sense like people are not talking about failure cases. So this is one of the reasons this uh new NSF grant that we have called programmable cloud lab, where the idea is not just to publish good things but also everything so that people can learn from what has not worked out, you know, in a more systematic way. Those are very critical to make an AI more high-fidelity model. So unless we have this from the community, whether it's soft matter or hard matter, uh, it's very hard to go and progress as a community. Yeah, in my opinion.

SPEAKER_02 1:40:30

Your thank you. Okay, everyone else, your turn now. Fabian?

SPEAKER_01 1:40:34

Yeah, I'm I'm coming at it from a different, a little bit different angle because I'm we're we're using cells to produce version, and and we uh when we use cells, we realize how little we know about what makes them tick and and what and and we have the potential to gather a lot of data, but we have no idea what to do with that data. And there is where I see a lot of potential of having a very quick uh like uh loop of of learning and improving. And if we can translate that into medical applications, because once we study cells in in the lab, we can study tissue in the lab, and we then we can start understanding better how tissues actually behave within the the organs themselves. I I see that as a as a huge potential, but we need a lot of data. And as as you as as Professor Chadou said, we also need no not only the success stories, we need the failure stories. That's the only way to to really teach AI to make it useful. Otherwise, if you only feed it, only the you know so all the published data is usually the success stories. Okay. We're missing a yeah, you know, the mountain of of failures that is you know if if it would be made public, we could really train models on on that and make them.

SPEAKER_02 1:41:52

Yeah, I think nobody wants to be the most published failure author in the world, although that could be highly important.

SPEAKER_00 1:42:00

No, it's such an interesting conversation because what is bad data, right? You can you know be a researcher and be married to your hypothesis with the hope that the result is what you've been predicting. As long as your test was properly controlled, the result might not be what you were expecting, but that doesn't mean that it's bad data. That's that's just data that did not match your hypothesis. But it's important to train the model exactly right.

SPEAKER_02 1:42:25

Andrew, what do you think?

SPEAKER_04 1:42:27

From a commercial perspective, right now, AI is is certainly reducing the available capital for physical products. At least in the medical space. With with all this emphasis on on dollars going into AI driven work, it is making it difficult for other companies to raise and therefore innovate in related spaces. So that it it is limiting in that regard. But also the AI police have to replace it.

SPEAKER_03 1:42:57

Well uh Scott, are you online still? I I am. Just enjoying this conversation. Yeah, and I think you know, a lot of the systems that we work with, right? So we're making co-polymers that are segmented. There they're there are these very complex structures and then layering that on top of of the 3D printed versions of those things. Um, you know, there's so much room for uh for customization in in modeling uh you know, to predict outcomes with this and the influence of of uh especially the implant location and all of the biomaterials that these uh that may interact with in situ. So uh, you know, I think it is a tall order for us, but but it is a noble one and ultimately to get where we where we need to be in terms of uh improving outcomes of healthcare. I mean, it it is uh critical for us all to work together, good models, bad models, good outcomes, bad outcomes, uh, you know, to achieve just that. So and it does start, you know, with the with the uh understanding of the materials, be able to predict the outcomes of of simple things like mechanics, it you know, it and and that begins that interacting cascade uh with the body to to elicit those certain outcomes. So I'm excited. Yeah, no uh no data is useless in science. That is a great comment in the chat, right? So uh you know, yeah, absolutely.

SPEAKER_02 1:44:30

I think Abraham, have you talked about AI yet? Uh don't think I don't think you did.

SPEAKER_06 1:44:35

No, just yeah, no, and it's something that I'm also working on it, but I mean it could be also very exciting, but also a bit scary. But what also I guess Fabian and also Scott said and also others that we definitely need also more data for the especially the biological and also the biology biomaterial interface, cell biomaterial interface, and but for the material design and like for specifically, I mean in my field for biomaterial design is scary, and because it can do it can do a lot. Uh it's just how to validate it into the more pure biological aspect and the interaction with the biomaterial.

SPEAKER_02 1:45:13

Absolutely. Craig, what about you?

unknown 1:45:16

Thank you.

SPEAKER_00 1:45:16

First off, Ibrahim, I had absolutely no idea that there's more bicycles than people in the Netherlands. That was an interesting takeaway, and I did fact-check it during your talk, and you were absolutely correct. Not that I doubt it, but that was uh quite uh interesting. Listen, we spoke about so many different topics here, right? And we could easily have an entire conversation just based on one aspect of this discussion. You know, I think I'm gonna come back to how we opened up the conversation, Jenny, right? You know, you host a number of different events like this over the years. I'm a materials guy, so I acknowledge that I'm biased when I'm about to say, but it comes down to the quality of the materials, it comes down to uh the performance of the materials, right? It comes down to understanding the application. That was something that came up earlier in our conversation with Fabian and Andrew, recognizing that a lot of the problems that we're all faced in our respective niche area, there's certainly significant overlap, but ultimately there's a unique aspect of what our particular application is. I especially like the fact that the conversation was wide-ranging, both for those in the audience that were researchers as well as those of us that are more on the industry side. So I think the takeaways is that for me, when I listen, or in this case, moderate a session like this, you know, my head is spinning with all sorts of ideas. It's obviously very exciting. And hopefully an event like this allows for those of us, whether it be speakers, CEOs of the of our respective companies, or in the case of Jenny and 3D Heels, hopefully it inspires conversations thereafter.

SPEAKER_02 1:46:48

Thank you, Craig, for example, excellent moderator.

Final Takeaways And Closing

SPEAKER_02 1:46:51

And then we ran over time. Thank you for everybody who have attended this event. This will be on demand for two weeks. And so you can invite others who might enjoy or benefit from it to watch. And otherwise, I'll see you next time. Thank you so much, everyone. Thank you. Thank you. Bye bye.