Episode 23 - Zoe Cournia: Precision medicine, designing allosteric drugs, and the role of an academic editor

In episode 23, Zoe and me discuss the current status and promises of drug design, a field where many things seem trivial but nothing is really simple. Zoe talks about the challenges that precision medicine is facing, and how it fits into the grand landscape of future therapies. We go through some success stories, and I ask Zoe how the recent additions to drug design workflows help them in everyday scientific practice, both in terms of physics-based and data-driven models. We then spend a moment talking about the opportunities and responsibilities of a journal editor, and what are the lessons Zoe learned from creating her startup.

Transcript

Milosz: 0:00
Welcome to Phase space invaders. It's your host Miłosz Wieczór here. Today we're releasing episode 23 and my guest is Zoe Cournia, director of research at the Biomedical Research Foundationat Academy of Athens. Zoe's main field of research is computer aided drug design and her work these days focuses on developing small molecule anti cancer drugs that can target specific cancer mutations. Besides having claimed an impressive list of awards and recognitions, Zoe also works as an academic editor for JCIM, the Journal of Chemical Information and Modeling, and use her entrepreneurial spirit to found a mobile app company focused on evidence based labeling of chemicals in food, just to give you how many scientific adventures she's embarked on. So we start with a general discussion of the current status and promises of drug design a field where many things seem trivial, but nothing is really simple. Zoe talks about the challenges that precision medicine is facing and how it fits into the grand landscape of future therapies. We go through some success stories and I ask Zoe how the recent additions to drug design workflows help them in everyday scientific practice, both in terms of physics based and data driven models. We'll spend a moment talking about the opportunities and responsibilities of a journal editor, and what are the lessons Zoe learned from creating her startup. As always, there's a ton of interesting insights coming from my guests. So hope you'll enjoy the conversation. Zoe Cournia, welcome to the podcast.

Zoe Cournia: 1:52
Thank you very much for inviting me, for giving me the opportunity to discuss today about computer aided drug design and related topics. And congratulations on launching this effort

Milosz: 2:05
oh, thanks hope it's useful for the community. So Zoe, I've talked here before with quite a few people who sometimes venture into drug design or drug discovery just as a subset of what they routinely do. But I think you were the first guest who primarily works on drug discovery you have done so for a good while with quite remarkable results. So, Perhaps at the risk of going into a territory where everyone has a strong opinion already. would you mind sharing your own perspective of where the computational side of drug design stands? And what is the biggest promise and challenge of the field as of now?

Zoe Cournia: 2:41
Yeah, thanks for the question. Of course, everybody's entitled to their opinion. Everybody has their opinions, uh, strong or not. So I'm just gonna speak, uh, about myself, today. I hope nobody will be offended. so I work in, not in general in drug discovery, but in computer aided drug design. So that's a specific part of, uh, drug discovery because as you know, the drug discovery has so many stages, uh, stages. up to the preclinical stage, that is very difficult to work on all of them. So, in my opinion, we are experiencing, really a transformation of computer aided drug design. And that's driven by new softwares, by new methods that are being developed and molecular simulation techniques. And of course we have, artificial intelligence while, it's not new, it's not a new method. We know that artificial intelligence techniques exist since the fifties. The Nobel Prize in physics was, uh, awarded on that as well as in chemistry. for listening. But although these are not new techniques, we now have more data that we can harness and we can apply these methods. And of course, we also have advances in high performance computing. So we have, more powerful computers. We have GPUs, quantum computing is starting to pick up. It's not there yet, but we hope that it will be there soon. So there are some promises of, computer aided drug discovery that I can discuss, for example, because of these advances that I just mentioned, it's Uh, now possible to screen billions of compounds virtually, which, uh, if you can, uh, post process the results, in a rational way can lead, to speeding up the drug discovery process compared to traditional wet lab experiments. we also have the predictive models that I discussed, for example, with artificial intelligence. We have, reliable data sets, for example, protein ligand interactions, bioactivity data, that can enable accurate predictions of drug properties, binding affinities, toxicity, ADME. Profiles and so on. Uh, we also have, personalized medicine. We know that, uh, this is a new era of, uh, medicine. It's called precision medicine. So with computational approaches, we can, uh, integrate, Many different omics data, genomic, proteomics, to design drugs. We can find better targets through these, processes and through specialized, computational techniques, we can, design molecules for specific mutants of, uh, these targets. We also have, uh, people are now starting to work a lot on novel drug modalities, for example, protein degraders, molecular glues, RNA based drugs, in general, novel therapeutics. We'll discuss a little more next. So that's that. It's quite exciting for me because, for example, PROTACs, uh, is something that, is recently emerging.

Milosz: 5:58
can you, introduce the concept to those of the listeners who maybe haven't heard

Zoe Cournia: 6:01
uh, PROTACs. So, PROTACs is a new drug modality. so basically PROTACs stands for Proteolysis Targeting Chimera. it's a molecule that, is being attached on a ubiquitin ligase. Uh, so one part is engaging an E3 ubiquitin ligase. And, Another part of the same molecule Jake. Binds to a target protein. So basically what is the,

Milosz: 6:32
You conjugate two

Zoe Cournia: 6:33
you call, yes, basically you, take the ubiquitin ligase takes the proteins for the gradation. So you attach one part of the molecule, this proteolysis targeting product onto the ligase. And the other part new, target and the protein that which you want to degrade, for example, a mutant in cancer.

Milosz: 6:55
hmm. Yes.

Zoe Cournia: 6:55
when PROTACs binds onto these two, proteins, it's signaled for degradation. yes. also we have, uh, you know, more structural insights. cryo EM, has experienced a renaissance lately, so we have more structural data and we have molecular dynamics simulations that are advancing their force fields, QMM methods, uh, and so on. And, uh, also in lead optimization, we have, Different free energy methods that can be applied for lead optimization that are, uh, quite robust. So, because now I'm going into the challenges, so for example, in my opinion, one of the biggest challenges in, uh, drug discovery is the lead optimization phase. It is the most costly phase, and obviously, optimization for binding affinity in terms of free energy of binding of a small molecule to a protein is very challenging as well as the ADME profile. So in my opinion, this is the biggest challenge. and we have tools, for example, free energy perturbation calculations can be used on a series of molecules, so very similar lead series. With that method, you know which one is preferred for binding to the target protein. However, these calculations are very costly. so when they need computational power, also molecular dynamics simulations need computational power. QM calculations remain resource intensive. So we still, while we have seen an advance in computational power, we still need more, who knows, perhaps with a quantum computer, it

Milosz: 8:44
sense that lead optimization is. Among the most challenging, right? Because screening you can do

Zoe Cournia: 8:50
Yes, it's a bottleneck.

Milosz: 8:52
scoring functions. But here you can, you have to employ those physical, very kind of

Zoe Cournia: 8:57
Exactly.

Milosz: 8:58
to see all the influences of small

Zoe Cournia: 9:00
Exactly, exactly. And I was about to say, so the next challenge is the accuracy of the simulations. Also, of docking, uh, scoring, functions. they are fast, but they're not so accurate. So this is a challenge. It's good enough to find the hit, but when it goes into lead optimization, you need more accurate simulations. And of course, that also goes into force field development, into going into more complex biological systems. for example, If in lead optimization you have an accurate method, free energy perturbation calculations, you have a good crystal structure, you do the calculation and it fails. It doesn't give the experimental result. Why? Maybe the parameters of the force field are not good. Or maybe there's a water molecule that entered and is stable in that pocket that doesn't allow for the full complexity of the system to be simulated. So there's the complexity of the biological system is another challenge that remains including protein flexibility solvent effects allosteric regulation. another challenge is the uncharted chemical space. We, as I mentioned, we are able now to screen billions of molecules, but the vast chemical space has not been exploited fully, in my opinion. So beyond going beyond known scaffolds, is a challenge. Perhaps artificial intelligence could help in that respect. And since I mentioned artificial intelligence, I could also speak about some challenges being faced there. One of the biggest challenges that we are also facing in my lab is data quality and bias. We have data, as I mentioned in the beginning, but we don't always have the data that is needed to make an efficient, uh, and robust algorithm. So data quality and the bias is one of the biggest challenges right now. Basically in machine learning and artificial intelligence, what you put in is what you get out. So garbage in, garbage out. If your data set, it all depends on the data set. If the data is not clean, if the data is not unbiased, and if the data is not trustworthy, You are not going to get trustworthy, result. So poorly curated or biased data can lead to misleading predictions. And for me, that's one of the biggest bottlenecks right now with artificial intelligence and drug discovery. yeah.

Milosz: 11:36
between different data sets, right? because maybe one method, is used in several data sets, but the details will make those uncomparable.

Zoe Cournia: 11:48
about future directions, but before I go into that, I would also like to say that for now, AI is more like a black box. It's difficult to understand why specific predictions are made. and also in my opinion, they don't provide physical insights. So, Just to conclude this challenges part. I don't think that molecular simulations or physics based methods are going to be replaced by artificial intelligence simply because in artificial intelligence, you mostly intrapolate. You don't really extrapolate and you cannot make new insights or You only. post process with the knowledge that you already have. You don't create new knowledge. So in my opinion, physics based simulations will still continue to, provide new insights and be used by our community.

Milosz: 12:43
Right, I think I've seen this, commentary recently about this based methods, right, that they mostly memorize the training set and then results that are similar to what was in the training

Zoe Cournia: 12:55
Exactly. And this is our experience. We have tried. Multiple, flavors of AI and it's always the same if the training set has not seen a property, it cannot reproduce it and it cannot predict it.

Milosz: 13:10
So we need explainability.

Zoe Cournia: 13:11
Yes, exactly.

Milosz: 13:12
Right. So let's jump then into, one of your specialties, which is precision medicine through allostery. if you care to explain the concept, what has been proven to work, what are the things that you hope will work soon? where is this subfield now

Zoe Cournia: 13:29
in my opinion, it's a very exciting field and it's a future of drug discovery, because, A leverages this concept of allosteric regulation to develop new molecules that are specific isoform specific, even isoform specific within a protein family. And they can be also mutant specific, so they can be targeted to a single point mutant. over a wild type of a protein, and these modulators that can, be developed through allostery. They can be used for targeted therapies to treat, uh, individual patient profiles. So let me explain what is allostery to my understanding. Allostery refers to the process by which a molecule would bind on one side of a protein, which is called the allosteric site. And then it uses a conformational change or a functional change at the distant site, the orthosteric site. So you have the active site, and then you, on the surface of the protein, you have other places which can influence the orthosteric site or the active site. by modulating a far site. So this approach provides unique advantages, because it's very specific. It's very controlled and addresses challenges, by traditional active site inhibitors. so for example, I can give you some advantages that are in my mind. so this allosteric sites. They are less conserved across protein families. For example, if you have protein kinase A and protein kinase B and they're very similar, they're doing the same reaction. The active site is very similar. but then these allosteric sites, they will not be similar across the same, uh, protein family. And that allows for creating modulators with greater specificity and reduced off target effects. Another important point is that these allosteric modulators can fine tune the protein activity. For example, you can achieve partial inhibition instead of complete inhibition. Or, they can be targeting specific, patient mutant profiles. Also, there was a study, several studies actually in the last five to seven years that show that allosteric modulators are less susceptible to resistance mutation compared to orthosteric drugs. So we all know the problem in cancer of drug resistance, so you take one drug. And then after a while, the cancer mutates and avoids this drug. Now, people have shown that if you treat combination therapies, the orthostatic inhibitor together with an allosteric inhibitor, the resistance profile is eliminated or limited. And, and that's very important. And, uh, yeah, that's, I mean, Exactly. And this, uh, concept has not really been exploited so much. So I think we are, with Alostery, we will have a new era in drug discovery. And for sure, we can exploit all targets with new chemical space because of these new pockets. And also we can target proteins that have been considered undruggable in the past. And I will give you examples because you'll ask me, for example, the most famous or recent example is the example of KRAS. It's a protein that's mutated in 30 percent of cancers. And, it was not possible to, drug it because there was one pocket, the orthostatic site, which had the picomolar substrate, GTP. So you couldn't really remove it because, you know, it's a picomolar. So nothing is better than that, obviously.


Zoe Cournia: 17:10
But researchers found, Kevan Shokat, in his lab they found an allosteric pocket where they designed sotorasib, and this sotorasib targets a specific mutant in cancer, it was not possible to do it before. Previously KRAS was undraggable, it was not possible to target it, but with KRAS, allostery, It was possible to launch a drug and FDA approved drug in the market in May 2021. So I think we're experiencing, I mean, I have more examples, but I'm sure we have more, uh, items to, to go.

Milosz: 17:49
Well, yeah, it's great to know that there are a few at least. So in this case, this doesn't rely on any mutation, right? So it's kind of when the patient has a mutation, no matter what the mutation is,

Zoe Cournia: 18:01
no, no. This is a specific, uh, it's a specific, uh, drug targeting. The Glycine 12, which becomes a cysteine. So it's a G 12C specific mutant. And that goes back to precision, to the concept of precision medicine. So with Aloster, you can really target specific mutants. So for the other mutation, G 12G of KRAS could be another mutant.

Milosz: 18:30
That was my next question. So how much can we rely on these repetitive mutations? I mean, how many cancers really feature a small set of well known mutations that are always the Rather than, I don't know, maybe a thousand of small mutations that never repeat.

Zoe Cournia: 18:47
Well, let me just finish the last question by saying that we have now a network, a European network that is called Alostery and Drug Discovery. And 11 companies and 13 academic partners are participating in this consortium. We are having our annual conference in, uh, April. 2025 in Athens. So if you type Aloster in drug discovery conference on Google, you will find that it's, between 9 and 11 April of 2025. I hope people who hear us, uh, will join us for the conference.

Milosz: 19:22
Uh, participants welcome. Yeah,

Zoe Cournia: 19:28
the next question?

Milosz: 19:30
the question was, how much are cancers driven by a small set of well known mutations versus maybe a huge number of random mutations that just mess up with something that

Zoe Cournia: 19:43
Yeah. okay. This is more a question for a biologist or a medical doctor. I am a chemist. So because I work at the biomedical research foundation in the Academy of Athens, I have exposure to geneticists and biologists. So I will just give you my uneducated experience here. so. Yes. I mean, cancer, because I think the question is on cancer. cancer is, uh, thousands of diseases. because often I get the question, when will we find the cure for cancer? Well, cancer is not one disease. It's thousands of diseases. So you have different subtypes. breast cancer, for example, has over 100 different subtypes that we have identified. So it is a multiple diseases. It's not just one disease. And in the specific, Types of cancers. there are obviously efforts going on. For example, the KRAS, mutants that we discussed, EGFR in non small cell cancer, BRAF, in melanoma. We have PA3 kinase again in breast cancer. So, roughly, I would say that maybe there are around 100, mutants that are being, targeted, or in the preclinical stage or in clinical trials or in the market. but this is very far from the actual mutations that exist in cancer. That's why I really think that gene and cell therapies that are coming up because cancer is a genetic disease. I mean, you can do so much with a small molecule. You can block the disease, but you don't cure it. You never cure it with a small molecule. You need to eliminate the mutation. And this is possible only with the gene therapy. Gene therapies, of course, are not available. But they will be. sure. We saw the advances on CRISPR. And we have data that show that gene therapy is possible. So you could replace with a viral vector. You could replace a specific gene with a correct gene. That would cure cancer. The small molecules only slow down the progression of the disease. new and new and new molecules are coming, people are changing their therapies and, you know, they have a better quality of life and a prolonged life, but you are never cured if you don't replace the gene. So I think the future lies in gene therapy as well as, you know, Cell therapies, you know, the CAR T cells for some indications, a few cancers,

Milosz: 22:23
So again, this for context, these are therapies, that rely on ex vivo treatment. Yes. If I remember correctly, and this CAR T therapies,

Zoe Cournia: 22:31
The CAR T therapies is for, yes, mostly for immunological cancers Well, basically they take your cells, and they reprogram them to find the cancer cells and kill them. So then they kill all the cancer cells. Which means that the mutation is not there in your body. So that's the concept discussion of the CAR T therapies. But obviously these are very costly. I mean, I'm not sure if the number is still correct, but I heard it's like 1 million dollars to get the therapy.

Milosz: 23:02
So that's part of the, Problem here, right? That sometimes we will have those almost bespoke therapies that will be made for a single patient. But then that means that a team of scientists is working on on an individual case. Do we also have effort in this like orphan disease, direction? Let's say those very rare diseases that can be maybe targeted by repurposing or, this is something that has less interest in in the field.

Zoe Cournia: 23:29
obviously, I mean, uh, specific, efforts are being made for the orphan diseases. and of course, repurposing is one way to find, new molecules using old modalities. and of course, yes, experimental, uh, repurposing of drugs. it's something that's going on and we saw it a lot with, COVID. So It is an approach that can be done.

Milosz: 23:55
So we probably need hundreds and thousands of scientists working on every possible branch of this problem, right? With so many mutations, so many diseases that rely on individual mutations and maybe each of them requiring some particular treatment. that's really the challenge of the whole field. So if you don't mind, sharing with us, how does your typical workflow look like if you approach and use, let's say, a new system, a new mutation, how would you approach this problem now versus maybe 10 or 15 years probably most of the people take their knowledge from.

Zoe Cournia: 24:33
Yeah in our lab, a typical workflow would start with allosteric or cryptic pocket identification. So, in the beginning, we need to find a new pocket if we don't have a pocket. But usually we work with challenging targets. So that means that the pocket is not there, or it's occupied by a high affinity substrate. So the first thing to do is to find the pocket. And how do we do this? we run, uh, typically molecular dynamics simulations, either vanilla or unbiased simulations, or by simulations, if we know that the conformational change is happening, whereby a The energy of the system towards this direction towards a big conformational change and that conformational change in the reveals pockets that were unseen before and also reveals how the protein is allosterically modulated. So, we screen with different tools. for pockets initially, and then we test whether these pockets are allosteric. What does it mean? It means we want to see whether the pocket that we have discovered communicates with the active site or other functional areas of the protein. in other words, whether there are networks connecting the two pockets. And there are a number of ways to do that, usually with cross correlation of, uh, residues. normal mode analysis. So there's many tools out there. so once we decided, okay, this is the pocket of interest and, you know, we check, uh, whether the pocket has available, uh, druggability. if it has good characteristics that would make the pocket druggable. We usually would run some simulations with a probe in the pocket to see how it evolves in time, whether there will be an opening or a closing. we also use recently, we are also using machine learning techniques. for understanding these allosterin proteins and for predicting if a pocket will be allosteric through data sets that we have gathered from the literature. So this is a project currently running in our lab. So identifying allostery through machine learning methods. So that's the first part to identify where is an allosteric pocket. So once we have the pocket, or it could be the active site. In some cases, the active site is still interesting. then we proceed with a normal virtual screening. workflow and although, you know, I say normal and one could say, okay, virtual screening, anybody can do the virtual screening and my experience. I see that, you know, not anyone can do the virtual screening. You need the. Deep chemical insights to understand the protein ligand interactions of what they would possibly be in reality and what you see on the screen. So you need to know your chemistry quite well, you need to know the hybridizations, You need to know whether a molecule would be cis or trans configurations, whether it would be an easy ester, the hydration that you see if you choose to put molecules and docking. So, there are a lot of parameters that are crucial there. So a good, virtual screening exercise. Would mean that the person knows what they're doing. And of course, we'll have the ensemble based method. because in visual screening, the target is not moving. You can have multiple structures to visual screen and then get the consensus. results from those structures. so this is the classic docking in our hands. It has, uh, success rates of around 20%. So if we screen, millions of molecules, we select 100, usually between, uh, 10 or 20, molecules will have micromolar affinities. and then if you have micromolar affinities, you go into lead optimization. Lead optimization, you can do it with de novo design, having ligand growing programs that would grow in different directions, your initial molecule and then Calculate the free energy of binding either with docking again, or if it's sufficiently different the molecule or with free energy perturbation calculations, or you can do scans with free energy perturbation methods. And, uh, lately, of course, we also use, uh, AI driven generative models. So these create novel molecular structures. Although we have seen, for example, diff doc is one, uh, method. but we have seen that, they are not always there yet. Sometimes, the molecules that are being predicted by these generative AI tools, they don't have the correct, druggability profile of a molecule. So sometimes the groups that appear, are far from those that exist in drugs that are currently on the market. so basically after that, if, when we identify, a hit, we verify it in, in, in vitro testing. We run lead optimization cycles. Then again iteratively going into experiments. And in the meantime, we also do molecular dynamics simulations. We have found molecular dynamics simulations to be a good predictor of the dynamic behavior of a molecule and whether it would be stable in the pocket. So we're routinely, I would say, we also do molecular dynamics simulations on the protein ligand binding. And something that is crucial on this step is the reproducibility. So you cannot do it only once. And see, oh, it's stable, so I'm fine. You have to do multiple replicas. I would say ten, ideally. But if you don't have the resources for ten, three, are good. So I would say this is, is the workflow that would follow.

Milosz: 30:45
curious, when you mentioned running simulations with the ligands, are the new neural network potential something that is going to help here? Or because I know that many of those strange or new molecules don't really parametrize well or require a huge effort to parametrize,

Zoe Cournia: 31:02
that's a, that's a really big issue. So every molecule is unique. It has unique chemical properties. So the parameters are not transferable directly from the force field. Because, okay, amino acids have been very well parametrized. The lipids have been very well parametrized. The water, but these don't change. They're always the same. The small molecules are always different. So efforts in the parametrization of, A small molecule is needed before you run this relation and we have a history in force field optimization. So we do take care of that part. So, in my opinion, when they are ready and robust, the new, machine learning potentials, they will be Definitely tremendously help speed up the calculation because okay, getting the parameters. We can also get them, uh, you know, manually, but this takes a lot of time. So if we could have them in a matter of seconds, this would be a game changer.

Milosz: 31:58
we had Alex MacKerell on the podcast, who is of the main people behind charm generalized force field. And that's probably one of the go to places to get parameters. Right. But they also give you those scores saying, Oh, this is very unreliable. You have to validate

Zoe Cournia: 32:14
Exactly.

Milosz: 32:15
it's never easy.

Zoe Cournia: 32:16
Exactly.

Milosz: 32:18
And then my other question was, how do you then once you're done with lead optimization, are you. incorporating this so exactly absorption, you know, distribution, metabolism, toxicity, excretion data into your, into

Zoe Cournia: 32:36
Yeah, absolutely. I mean, that's critical. That's a critical part of a practical approach to drug discovery. So these ADMET properties, they can determine whether a compound can become a viable drug. So one cannot overlook them. Binding affinity is of course very crucial parameter in drug design, but ADME is the one to determine whether a compound can become a drug. So these properties, have to be integrated in the drug design pipeline, both through predictive models and experimental validation and so we have not worked with creating a predictive machine learning model. Although the data I would say is there, it's possible to do it. And of course there are tools, Deep Chem, ADME Lab, Swiss ADME, there are many tools out there that we use. And, you can calculate, the half life probability, hepatotoxicity, and other properties also that we calculate that don't need machine learning, like molecular weight, then solubility log p. So we definitely take a look at those. so far, my experience is that also these models are not there yet, but they are improving, especially the commercial solutions, the ones that you pay for. they are reaching a very good level of validity. So I would say, that's definitely a critical task in drug discovery.

Milosz: 34:02
It feels like drug discovery is one of those fields that are kind of driven by big capital, Maybe more than on the methods on method development with AlphaFold and things like that, but the big money is probably exactly in the drug discovery drug design side. and to move away from the drug design side a bit. What are your experiences as the academic editor of JCIM, the Journal of Chemical Information Modeling? And also as an entrepreneur, right? This is also something you do. are those of the, let's say, academic enterprise different from the standard life of a PI if you share a few

Zoe Cournia: 34:45
absolutely. Um, first of all, the editorial work, I'm really grateful for the trust of our editor in chief, Kenny Merz. I found it very, extremely exciting. I learned so much, but I could also contribute. a few things that I think are important, for example, data reproducibility, principles, open science, advancing women in science, especially in computational chemistry, some, uh, you know, practical considerations of, dynamic simulations, also machine learning. So a very big part that I find rewarding, apart from the cost, typical assigning manuscripts, reviewing manuscripts and so on is the policy making that you can do. And, uh, we have published, few, editorials. For example, uh, one is on, method and data sharing and reproducibility of scientific results. And I think that, JCIM is a pioneer. in this area, because we before we accept the paper to go for review, we ask the authors to provide all the data that were used. To generate the results, parameters, topology files, scripts, analysis scripts, the data sets for training. Everything that is needed to reproduce the study should be part of the supporting information or should be uploaded in an open source repository. That's really crucial, in my opinion, because the era of, data available upon request is hopefully behind us is we need to make the, you know, how, how many billions of euros are spent in duplicating research efforts because the data is not there. So every year, so we need to go over this and that's why I'm really grateful to the vision of Kenny Merz, who enabled this policy in JCIM now all The data from all the papers that are needed to reproduce the paper, the data in the paper are available. We don't ask for the outputs or trajectories. Okay, this is on the person if they want to make them available. But to generate the data, all the inputs need to be there. So that was one really important contribution for me. So open data and open science. And in terms of, women in, science, we also published two special issues. which for me, you know, there are two sides of the coin. One says, oh, why are women special? Why should you have, special issue on women in computational chemistry? But I think, I mean, it has been shown that there are inequalities. So, we really Thoughts that we could, uh, publish this special issue to build support networks to make aware of the issues that are there so that we promote more, for example, women speakers, moderators and conferencing panels, to ensure that general balance on important decision making committees is there. You know, if we witness unfair comments, we are all, saying something about it, recognizing implicit bias. for example, to ask our governments to offer funding for professional reentry after a woman gives birth. you know, there are so many things, uh, that, uh, feel, uh, that we have some work to do. that I thought it was important.

Milosz: 38:24
It feels that the computational side was always. Particularly biased towards men. So yeah, that's a

Zoe Cournia: 38:29
And that's what the statistics say. I mean, I don't say this, but it's true. Because if you see the statistics when you are at the PhD level, you are 50 50. 50 percent male, 50 percent female. But when you go into the professor rank, 26th professor is women. And 80 percent, 74 percent is male. So clearly there's something going on there. And I think it has to do a lot with, when you are about to become a professor, this is on your childbearing years. You want to have children, you want to grow your family, and typically this stalls you. from being promoted to a higher rank.

Milosz: 39:10
Right. We just had this conversation with Lucie Delemotte who has just become a mother last year and yeah, she made this point that, once you have tenure, it's really easy, but then biologically it can be also hard to, meet those,

Zoe Cournia: 39:22
Yeah, absolutely. And, we don't count also the, because you know, when we are in the academic track, we are striving very hard to go close to the tenure in any, not only in academic track, but also in the professional industrial track. And you may be unsuccessful in creating a family. You may have miscarriages, infertility, unsuccessful adoptions, and these take a big toll on the women's psychology, and these stall, the progress. So that's why I think we have low percentages of women in high professional ranks. But with these special issues, what we hoped to do was to create a network. So we asked all the authors, that were contributing to this special issue to write that they support advancing women in science. In this way, if one opens this issue, they know that, okay, these authors are supportive. I could ask them to mentor me, for example. So this is what we wanted to build with these two special issues, to find people who are believers of advancing, only women, but in general, to promote equality in all aspects and to build a network of these people that are believers in equalities. So that, that was very important for me. so this policy making, let's say. Also for some guidelines or best practices molecular dynamics simulations. I mentioned it before. I think reproducibility is key. So we published this editorial on best practices on performing molecular dynamics simulations. Where we say that, okay, look, you have to do a triplicates of your simulation. One simulation is not enough. so that was important for me. And apart from those policymaking, uh stuff that I mentioned, there's also some other insights for me personally. So I learned, that is important in the article to have clarity and focus. to show the novelty. often, the comments I receive from the reviewers is that the manuscript is not novel. But it was not that it was not novel. The manuscript was novel, but the authors did not, pronounce the novelty. enough to be understood by the reviewer. So that's important to show the novelty of the work, to validate the work, uh, to make conclusions that, uh, are supported by the data. And also I learned how to provide constructive feedback because what's important is to make a scientific contribution to the community that is valid and will be used. So we need to provide reviews that are constructive, are respectful and solution oriented and we already talked about the data. That's another important point data sharing and transparency that enables reproducibility and open science. And, uh, yes, I think I already said a lot about that. I, I found it very, really rewarding. I learned a lot of things also about writing my own manuscripts. But I also learned that, you know, the editors their job is really to facilitate that new innovations are coming out into the scientific community that are valid and that they don't have plagiarized that the discoveries are solid and have novelty. So that is like a pre screening tool for the community. 

Milosz: 43:06
Out of curiosity, how many papers go through your hands every day?

Zoe Cournia: 43:10
it's not every day. I would say maybe five. Yeah. Five or less.

Milosz: 43:15
I wanted to connect it to the point about highlighting the novelty, right? That if an editor assigns a given number of minutes to every paper, because of course you cannot spend all your day just reading the papers all the

Zoe Cournia: 43:27
Yeah. I mean,

Milosz: 43:28
Those things have to be highlighted, have

Zoe Cournia: 43:33
I don't do the review myself. I, I sign, the manuscript reviewers. So.

Milosz: 43:39
you decide whether it's interesting enough, I guess, it part of the job

Zoe Cournia: 43:44
very few manuscripts, very few manuscripts that come into JCIM are not going into review. I will give you some examples. if you have a drug design paper that comes in, but there's no experimental validation, this will not go to review because it's just a putative computational hit. obviously if it's just computational, we don't have experimental validation, we cannot send it for review because The result is questionable. so only such extreme cases will not go to review. am not the one to judge the novelty or

Milosz: 44:21
Yeah, I think I think it's useful for the people to hear what the rough rules of thumb are from the

Zoe Cournia: 44:26
yeah, I can give you more examples. Uh, for example, if, there is a paper coming in that explores, particular midazole salt and its ionic, solvation cell. that is very narrow. This goes into physical chemistry. All right. So there is a scope for JCIM. So basically, one of the best guidelines to remember is that when you are submitting to a particular journal, you have to read the aims and scope of the journal. For example, JCIM, the Journal of Chemical Information and Modeling, is a journal that wants to have advances in applications of molecular simulations, developments in multi scale modeling, artificial intelligence, machine learning models to chemical and biological data. So there is a list of what we are interested in, QMMM applications and methods. Now, if it is a very narrow system that focuses on the physical physical chemistry aspect of the system, then the journal of physical chemistry is more appropriate. So, people should also read the scope, the author should read the scope

Milosz: 45:40
Right.

Zoe Cournia: 45:40
of the journal, of each journal.

Milosz: 45:43
A good point. Sure.

Zoe Cournia: 45:44
Yeah, you also asked me about the entrepreneurship part. So we have developed small startup company. It's called Ingridio. It is a mobile phone application that helps consumers understand dangers from chemicals. So basically, it's a mobile phone application. It's available on Android and iPhones. It's free to download. what it does is that it has a database, it has collected the peer reviewed literature from PubMed and PubChem. And, we have collected also the database of chemicals that are found in food and cosmetics ingredients. so we have created a machine learning algorithm using natural language processing that can identify causal relationships between ingredients of food and cosmetics and, adverse effects on human health. So, let's say That, one product contains one specific chemical. Then, the user with the app takes a picture of the bottle. the algorithm determines the chemical ingredients, looks into the database, finds them. And then, We assign a specific score of whether this ingredient causes adverse effects on human health based on the peer reviewed literature, based on, scientifically, uh, reviewed data. And that comes up to the user as a color coded, score, which says, uh, okay, if it's, if an ingredient is red, it means probably it's not good for you. And there's an explanation about if an ingredient is green. then it means it's not harmful. and the key insights that I think I got from this process is, first that when you are going into an entrepreneurial adventure, what is important is to solve a problem that people have a real problem. And this idea came from my friends who were like alarmed by news that they heard that some foods cause cancer or diabetes and they couldn't read the label because, okay, they are not chemists. So I could hear all the time that we don't know what we are consuming. So this was a problem that my friends had. And so, and I also had this issue, although I'm a chemist, I don't know all the, E ingredients, E 101. I cannot remember it by heart. So we made this application for people to use in the supermarkets right on the go. So it's really important to target a real life problem. And it's the most important thing that I learned is very important is not the money. Because money you can find, or if your product is good, you will generate the revenue. It's to get a good team of people that will work with you and to hear your customers. Hear what they have to say and what they need. and the market need. That's the most important thing. The money is, uh, of course a bottleneck, but it will be if your idea and product is good, you will find the money. and there's also a lot of support for SMEs. You know, if European funding opportunities, there's so many incubators these times, because I also, I'm not business developer, right? I haven't studied finance or economics, so I went into accelerators. incubators and learn the tricks of the trade.

Milosz: 49:23
Right. I see. I think it's also a great connection, right? Between the academia and the public in this way that results can somehow be transferred into what the people see and hear.

Zoe Cournia: 49:37
Definitely.

Milosz: 49:39
Okay. I think we've got a lot of insights from you. Thank you again. Zoe Cournia. for sharing your experiences and insights and being on the

Zoe Cournia: 49:48
Thank you for inviting me.

Milosz: 49:50
Have a great day.

Zoe Cournia: 49:51
Bye.

49:52
Thank you for listening. See you in the next episode of Face Space Invaders.