The Future of Genomics: AI, Synthetic Data & Human Expertise

In this episode of Pulse, Aldo sits down with Dr. Vinh Tran, Lead Bioinformatician at NIOME & Researcher at the Goethe University in Frankfurt. 

Together we explore how AI and synthetic data are reshaping the future of genomics and life sciences.

They discuss why data quality remains one of the biggest challenges in bioinformatics, how synthetic genomic data can accelerate research while protecting privacy, and where AI has the greatest potential as well as the greatest limitation. 

Vinh also shares his journey into computational biology and why human expertise remains essential in scientific discovery.

If you are interested in AI, genomics, or the future of medicine - then this is a great Episode for you!

Sincerely hope you’ll enjoy this conversation with Dr. Vinh Tran of NIOME. 

More on NIOME can be found at: https://niome.genomes.io

Watch this episode on YouTube: https://www.youtube.com/@AlphaWireHQ

This episode was brought together by AlphaWire: https://alphawire.xyz/

Transcript

SPEAKER_01 0:00

We've got this 24 challenges roadmap set out, and we've done a first one on the CFTR gene looking at cystic fibrosis, right? And the results already were incredible. Out of 190,000 base pairs, we identified, I think, 6,000 genes that were relevant to the CFTR gene, and I think 1,500 of them were directly relevant to cystic fibrosis. That is massive compute already, right? Those are great numbers. So, you know, such good news there. Welcome to Pulse by AlphaWire, the podcast where science and education meet cutting-edge technology and artificial intelligence. My name is Alderap, and each week I sit down with innovators, thinkers, and doers who are working to change our world for the better. Together we explore their journeys, uncover the lessons they learned, and take entrepreneurial pulse that drives them on their path to success. In this episode of POLS, I sit down with Dr. Vin Tron, lead bioinformatician at NIOM and researcher at the Goethe University in Frankfurt. Together we explore how AI and synthetic data are reshaping the future of genomics and life sciences. We discuss why data quality remains one of the biggest challenges in bioinformatics, how synthetic genomic data can accelerate research while protecting privacy, and where AI has the greatest potential as well as the greatest limitation. Vin also shares his journey into computational biology and why human expertise remains essential in scientific discovery. If you're interested in AI, genomics, or the future of medicine, then this is a great episode for you. I sincerely hope you'll enjoy my conversation with Dr. Vintran of NEOM. Yes, and it is a wonderful Friday morning here in the United Kingdom. I am very happy to be sitting here at home in my study with a great uh colleague. Uh, a colleague uh who I'm working with in my biotechnology company, genomes.io, also known as NEOME. And this conversation is all going to be about artificial intelligence and bioinformatics. And I have the great honor to be talking about those topics with Dr. Vin Tran. Good morning, Vin. How are you?

SPEAKER_00 2:38

Good morning, Anno. Thanks. I'm very good. Thank you.

SPEAKER_01 2:41

Vin, I wanted you on the show. Um you were hesitant. You are an extremely clever, great guy. You were hesitant because you are a Vietnamese origin, moved a long time ago to Germany, and as a third language you speak English. You were a little bit insecure about your English, but I think it is very good. But we do take note, and I hope you will take note as well, dear listener, that I think it's very brave what Vin is doing here today. So, Vin, a very warm welcome to you. And I would like to know a lot more about the world that you operate in. So let's dive right in. What got you into bioinformatics?

SPEAKER_00 3:24

Thank you, Andrew. Also I uh initially studied uh biotechnologies uh during my bachelor's degrees in Vietnam, and as why I enjoy learning about biologies, as I became increasingly interested in working with the biodata and computational approaches as rather than just performing great led experiments. And what fascinated me was the idea that we could assimilate biological processes, analyze the last data sets, and even generate new insights using just the computers. So I saw how the computational methods could make the biological research more efficient by reducing both the time and the cost that we require to test the hypothesis experimentally. So therefore, during the final years of my bachelor's degrees, I chose bioinformatics as the focus of my thesis project. And thus experience confirmed that this is the real field that I wanted to go in. And it allowed me to combine my interest in biology with problem solving, programming, and data analysis. And after that, I moved to Germany to continue my education with a master's degree and later a PhD in bioinformatics. And since then, I have never looked back. What's still exciting to me today is that bioinformatics continuously evolved alongside with the advance in computing, genomics, and AIs. So creating new opportunities to answer the biological questions that were previously impossible to address.

SPEAKER_01 5:01

Nice. Wow, that's such an amazing scope, such an amazing journey. What do you consider to be actually the biggest limitation of bioinformatics? Because for me it sounds like sky is the limit, but are there any limitations to bioinformatics, you know, as far as you could see?

SPEAKER_00 5:18

Yes, of course, everything has limitations, right? So um one of the biggest limitations in bioinformatics, as I can see, is the data quality. So my supervisor often says, so garbage in, garbage out, and I think as I just summarized a challenge very well. Um so most of the bioinformatics methods relies heavily on existing data, and whether we are identifying disease-associated genes or predicting protein functions or training machine learning models, the conclusions we draw are only as reliable as the data we use. So in many cases, also bioinformatics is basically learning the pattern from previously observed data and applying these pattern to new datasets. So if training data are incomplete, bias, noisy, or poorly annotated, then the resulting prediction can be inaccurate or misleading. And so another challenge is that the biological systems are very complex. So even when we have a large dataset, so they may not fully capture the diversity of the biological variations found in nature or in our human populations. So this is particularly problematic for, for example, rare diseases or for underrepresented populations, as well as high-quality data can be very limited. And for these reasons, I think generating high qualities and well-corrected and representative data sets still remains one of the most important priorities in bioinformatics. And so better data often leads to better models and more reliable in biological insights.

SPEAKER_01 7:07

I like what you say: garbage in, garbage out, and the entire data processing end. So the quality still needs to be good. And I can imagine there still needs to be humans to kind of make sure that the data is solid. What did you think when you were approached to kind of join Team Neome and Genomes.io? So working on this big project in artificial intelligence, uh, looking at synthetic genomic data. So, what were your thoughts when we approached you?

SPEAKER_00 7:35

I was super scientists as when I was approached by you about zoning neoms as a lift bioinformatics. So, what inmately caused to my extension was that neon business of building a decentralized infrastructure for generating high-quality synthetic biological data. So throughout my research career, I have encountered the same challenge. So the success of modal computational and II methods depends heavily on the availabilities of large and high quality data sets. And so today I have seen that many researchers want to apply state-of-the-art machine learning approaches to biological and medical problems, but the availability of suitable training data often becomes the bottleneck in this process. And this is particularly true for areas such as rare diseases, specialized clinical studies, and any other domains where collecting the last amount of real work data is difficult or expensive or limited by privacy concerns. So therefore, um I believe synthetic data has a very high potential to help address these challenges by expanding the available data sets while still preserving the important biological characteristics. So new goal of uh creating purposely accessible and high-quality synthetic data that can support the scientific research and AI development is therefore for me as both timely and highly relevant. And what interests me most is the opportunity to contribute to projects that sit at the intersection of Xenomics, AI, and distributed computing. And it is not only significantly interesting, but it also has the potential to create tools and resources that benefit the broader research communities. So that's why I'm very happy that I could enjoy NEOMS and to be part of your team.

SPEAKER_01 9:44

Wow, well that's great. There's a lot in your answer. We just also had the NEOME Summit, as you are aware. We had loads of people talking about artificial intelligence and unlocking the life sciences with artificial intelligence. Where do you think that AI can be a game changer in the life sciences?

SPEAKER_00 10:02

I believe AI can be a game changer in areas where we as humans need to extract patterns from extremely large and complex data sets. So as you already know, in biologies and medicines, we are now generating data at a scale that would be impossible for researchers to analyze them manually. So in this case, AI can have to identify the hidden relationships between genes, proteins, diseases, and biological pathways as much faster than the traditional approaches. So for example, as we have seen, uh AI has already been proven its values in like protein structural positions, also genomics uh analysis, drug discoveries, and uh medical images. So it can speed up the hypothesis generations, also have prioritize experiments and support researchers in making sense of a huge amount of information. So for me, also AI can act as a research assistant for scientists, so it allows them to spend less time on repetitive analysis and more time on interpretation activities and designing as a new experiment. So I think one of the most exciting opportunities is combining AI with high-quality biodata to better understanding the human biologies and further improve, for example, healthcare outcomes.

SPEAKER_01 11:35

Kind of on a lot of levels it can be applied. Let me now ask it the other way around. Where do you think AI is hype? Where do you think it won't deliver as people say it would? It cannot be a magic wand everywhere. So where do you think it it kind of won't do as as we think it will?

SPEAKER_00 11:53

Yeah, good question. So on those, uh AI is very powerful, um, but um in my opinion, I do not believe that this is uh suitable for human understanding and critical thinking. So um for example in biologies, so I models can identify correlations and patterns, but they do not automatically provide biological explanation. So scientific discoveries still require us as researchers to formulate the hypothesis, also decide experiments, interpret the results, and evaluate whether the finding makes biological sense or not. Another is uh II is also limited by the data it's learned from. So if the underlying data are biased, also incomplete or inaccurate, the II system still can produce misleading conclusions with a very high degree of confidence. So this is very dangerous. And for those reasons, also I see AI as a powerful tool rather than a replacement for scientists. So the most successful application will likely be those where the human expertise and AI capabilities complement each other. So, in my opinion, also human judgment, domain knowledge, and scientific reasoning remain very essential.

SPEAKER_01 13:17

I mean, this is the word that loads of people use, is hallucinating, right? So AI is very good at being very convinced in hallucinating, and and you're almost ending you end up having an argument with AI saying, well, no, that's not what it is, and AI will continue to say, no, no, that is what it is. Maybe because the input was wrong or whatever it is, but that hallucination is a dangerous thing, right? And specifically in the life sciences. I just wanted to share an observation also of the of the summit that we held, whereby loads of people were mainly concerned with, as you say, the garbage in, garbage out metaphor, and also tied that back to safety of information. Because giving your data to artificial intelligence or using that for AI modeling or AI training, you don't necessarily know where it's going to end up if you use with closed AI models, so to say. That also prevents loads of renowned labs and research universities and and whatever to take a step back when it comes to using AI. Now, I think that's good news for us because we focus on synthetic data, so it's never real human information, it's never real biodata, but it is as good as, right? So it's it's it's kind of like how do you train AI in a safe, secure, private way without any risk coming back to you. I think it's too early days to do that with real biodata. You don't know how it's gonna come back to hurt you. And specifically in this day and age where where cyber hacking and data breaches are at an all-time high, I think we need to be very careful. So I think that was one of the things that kind of from all the companies and organizations that were there, we kind of heard back of okay, yeah, we need to be cautious and careful. We can't just get on the AI bandwagon without thinking it through because you just you've got one opportunity to do it right. So let's take it essentially. Yeah, of course. So interesting, Vin. You you obviously have so many great ideas around this, and we're we're so happy to have you being part of the team. We've got this 24 challenges roadmap set out, and we've done a first one on the CFTR gene looking at cystic fibrosis, right? And the results already were incredible. Out of 190,000 base pairs, we identified, I think, 6,000 uh genes that were relevant to the CFTR gene, and I think 1,500 of them were directly relevant to cystic fibrosis. That is massive compute already, right? Those are great numbers, numbers that we would never have been able to achieve if we were just working with our infrastructure where we are in the thousands but not 190,000, and so on and so forth. So, you know, such good news there. And we've got so many other challenges coming. So it's been really great to dive into this with you. Thank you so much, uh Vin. I'm gonna dive into my two very last questions because I appreciate you're busy. We've got loads to crack on with, and I think this gives our listeners a good insight into NEOME and Genomes.io. So for my two very last questions, and I ask everyone uh uh on this show, I'm a little bit of an amateur researcher into morning rituals. Do you have a morning ritual? And so would you care to share?

SPEAKER_00 16:42

My uh mornings are usually quite busy because I have a young son. So I know normally start the days by having breakfast with him and then helping him uh get ready for school. And after dropping him off, I thought I drive to begin my work day. So um I actually enjoy this routing because it provides a clear transition between family life and work and give me time to organize my thoughts before arriving. And although every morning can be a little bit different, but uh spending time with my family is an important part of how I start a late.

SPEAKER_01 17:19

Wonderful, that's beautiful. Yeah, and I also enjoy my breakfast with my daughter and then taking her to school. It's very wholesome before you kick off with today. And we should enjoy that because blink and it's over, right? There will be out of the house, they will be grown and gone, and they won't need you soon.

SPEAKER_00 17:35

So, like, we should enjoy the time because yeah.

SPEAKER_01 17:40

And then for my very last question, which reading has inspired you on your journey? For those so for those people listening and kind of you know who are intrigued in bioinformatics, or maybe you as a person, has there been any reading that kind of said, like, well, this is really kind of important to me and and this is something for you to get into? And it could be anything, could be an article, could be a book, could be, you know, maybe a movie that inspired you. Is there anything that you would care to share?

SPEAKER_00 18:07

So uh as a scientist, also I I I read a vast amount of scientific papers, uh, but I wouldn't say they are what inspired me the most personally. So I usually read the papers because it is necessary for my work and help me to stay up to date with the development in the field. So actually, the book that have inspired me the most are actually quite different. Also I um enjoy reading novels and stories about childhood's uh family life and personal growth. Also, in particular, I like the book by a Vietnamese author called uh Wing San. His story often captures the experience, emotions, and memories of the Child Su in a very authentic way. And uh reading this book reminds me of my own time in Vietnam and helped me to stay connected to my roots despite living abroad for many years. So they also provide me a well balance to the highly uh analytical nature of scientific creatures and remind me of the importance of human experience and personal stories. Yeah.

SPEAKER_01 19:16

Wow. Do you know the title of the book?

SPEAKER_00 19:19

There are many, so I they so he has a a bunch of books, so I have here ten or even more.

SPEAKER_01 19:32

Okay. Well, I mean, it sounds amazing. I don't know if it ever got translated into English, but uh but thank you for sharing that. And and it's very nice that you like to read about something non-scientific, you know, and that gives you great insights. It's been great sharing a little bit of these insights with you and exchanging insights, I should say, on Genomes.io and NEOM. Uh, we're very lucky to have you in the team. Vin, thank you so much for everything that you do, and thank you for talking us uh a little bit through kind of your end of uh Genomes.io and Neome. Thanks so much, Dr. Vin Chan.

SPEAKER_00 20:09

Thank you, Andrew, for the opportunity to share my perspective and also looking forward to seeing how this field evolves in the years ahead. Thank you.

SPEAKER_01 20:21

You've listened to Pulse by Alpha Wire, produced by Natalie Piles and Amela Faisal, with great music, The Optimist, written by Holly Hamill, performed and produced by Alo. Episodes hosted weekly by me, Aldo DePunk.