Simon Burns: Alex, thank you for joining us today.
Alex Zhavoronkov: Thanks for having me.
Simon Burns: Let’s do a quick round of introductions. I’m Simon, co-founder and CEO of Vial. We’re next generation CRO building dramatically faster and cheaper trials powered by technology. Alex your, reputation precedes you. But for the audience, tell us a little bit more about yourself and how you started Insilico.
Alex Zhavoronkov: So, sure. My name is Alex Zhavoronkov. I’m the CEO and founder of Insilico Medicine. Actually, I’m a co-CEO of Insilico, so we’ve got two CEOs, it’s a long story. And before I started Insilico, I was in semiconductor industry, and my background is in computer science.
Around 2004, 2005, I decided to switch from semiconductors to biotech, because I think that the most important challenge and problem in life is aging and the age associated diseases. So, whatever I do is related in one way or another to aging research and addressing age associated diseases. So, I did my grad work at MSU. I did my grad work at Johns Hopkins.
And I worked for a number of biotech companies mostly related to aging research. And in 2013, 2014 during the renaissance of deep learning and the true revolution in deep learning when declaring system has started outperforming humans in image recognition, voice recognition, text recognition.
I realized that my business that are powering deep neural networks already had very substantial datasets cleaned up and ready for machine learning. And we started experimenting with those datasets utilizing deep learning techniques, originally record target application and then for generative chemistry.
And in 2014, I started Insilico originally at the emerging technology centers on campus from the Johns Hopkins University, on… so, I moved to Baltimore for a while. And now, we’re scaled globally. So, now, we are a truly global company with R&D centers in nine countries or regions spanning pretty much every time zone.
So, right now, I’m calling from Shanghai, China. I know it’s 1:00 A.M. and my next call is in 25 minutes. So we are trying to be omnipresent and trying to play in multiple areas of science technology, but also in many regions.
Simon Burns: It, feels like we’re, in a moment right now. This is the computational bio moment. And there’s a lot of contributing factors, I think, people are talking about of accessibility of, machine learning and AI in general, just that this pace advancements there.
There’s talk of just scale of compute available if the cloud is enabling this, new explosion in computational bio, there’s also talk of the CROs. It’s the preclinical CROs whereas WuXi or others that are enabling this kind of innovation.
Do you feel like this is the moment for computational bio? And, if so why, is that? What are the factors leading to this being the moment?
Alex Zhavoronkov: So, sure. I think that around 2014 to 2016 that was the true moment of what comp bio and comp chem utilizing new approaches, specifically deep, learning, deep neural networks. And that’s when the expertise in this field was very scarce.
Nowadays, high school students are doing target discovery using deep neural networks. I can perform pretty much at the same level as us in 2014. And today many of the technologies that we’ve been developing around 2014 to 2016 are now validated using experimental data and some are even in human clinical trials.
So, right now, it’s the moment for the companies that started again, around the time of the deep learning revolution and expanded into their own pipelines. So, started utilizing these technologies to develop their own pipelines and validate it to the level where they could go human.
So, now those companies that achieved this and also managed to fundraise successfully they are the major consolidator is of the market because there are many startups out there that actually have reasonably good tech but, they don’t have the level of validation that pharma needs in order to partner or investors need in order to invest.
And we are right now at the moment of the massive consolidation of the industry. I think that right now it is pretty much like 2001 in the .com era where the bubble has burst, right, So you’ve got major[00:05:00] collapse and the winter in the early-stage startups.
But the companies that manage to grow they are growing very fast, right. I think we’re having more abundance. And right now, it’s really the, time of growth. And of course, you’re absolutely right so the CROs are fueling this growth.
So, especially a company that managed to establish a very fluid and very operationally effective way to collaborate with CROs. So, you can actually distribute the workload between the many CROs and also parallelize.
And at the same time achieve certain level of redundancy because as may have one experimental result, another CRO may have another. And it’s a very good idea to have redundancy and to, expect one experiment in two different labs to get higher confidence and also generate a little bit more data.
And the availability of those CROs is allowing companies like ours AI powered drug discovery companies to move much quicker and distribute the workloads geographically, but also with them on multiple CROs that provide you with the level of infrastructure that you would not be able to achieve internally.
And of course, the availability of data has increased. And the availability of talent, as I mentioned. And many of the algorithms that we’ve been developing in the early days have now been commoditized. And some turned into ready to use software.
For example, Insilico, in addition to our own drug discovery we make the tools that we use internally available for the entire industry. So, you can pick up PandaOmics plot from software to service, and start discovering novel targets using many, different methods.
We have integrated more than 60 drug discovery philosophies, target discovery philosophies on the tool, and it’s reasonably inexpensive, you can put it on a credit card and big And Chemistry42 is the system which allows you to very rapidly generate small molecules with the desired properties.
It’s also ubiquitously deployed. So every pharma substantial expertise in AI as deployed Chemistry42 and… are now using it as a tool, multiple pilots but was a ready to use product. And the level of validation of those technologies has exceeded my wildest dreams when we just started Insilico.
And of course, that allows us to grow and the entire industry to grow. So, I think that again, we’re at the point of no return. For some of the companies we see the Googles and the Amazons and the Microsoft emerging out of this Cambrian explosion of small companies.
And most likely, those companies will continue to grow. Even though right now, as we’re in the biotic winter, so budget is sort of scarcest, companies are going down many companies are collapsing.
But I think the probably early 2024 end of 2023, we’re going to see a major recovery in this industry. And again, compile companies that validated and managed to deliver clinical stage assets are going to be the motherships of this industry.
Simon Burns: I ask you about compounds synthesis, the timelines, the process just synthesize ability as a key parameter and in a lot of small molecule drug discovery has been a key focus. Many, people point to that as the, part of the DMTA cycle that slows the drug discovery process is, a key bottleneck.
What’s your sense of, that statement? Is, the compound synthesis timeline really big… bottleneck? And what are you excited about in terms of new approaches whether it’s automated synthesis or, other approaches.
Alex Zhavoronkov: So, of course, synthesis as a service came available for many CROs. And also, now, we see very substantial level of automation in that area. And it really sped up the design or analyze cycles on the DMTAs pretty dramatically.
And also, think synthesis has gone down in price slightly recently. And new algorithms also, we are developing some and are some algorithms available from other vendors. For example Mercks Cynthia is pretty good.
And many other technologies are available that allow you for pretty effective and efficient retrosynthesis planning. So, you can land the, synthetic route more efficiently using AI. And with generative chemistry, you can use synthetic accessibility as a generation condition.
So, they generate with synthetic accessibility in mind. And also filter out the molecules that are being generated using synthetic accessibility predictors. And then, for those molecules that you, prioritize, you can do synthetic crowd planning.
And again, utilizing advanced CROs, you can very rapidly and test. So, this bottleneck is becoming less of a bottleneck. And the cycles are becoming shorter. But I think that the predominant reason, the primary reason why these cycles are also becoming shorter is that we don’t need to synthesize as [00:10:00] many molecules.
So, right now, at Insilico, for the first cycle, when we are trying to identify hits, usually synthesize 15 reasonably potent… may take maybe a couple months. And then the [inaudible 00:17:01] would take you maybe two or three rounds of synthesis and test.
And for every cycle, we will use 50 molecules, sometimes a little bit more if target is difficult. And the cycle will take you about four weeks that’s including in vivo testing. So the, speed has increased dramatically.
And also the number of molecules that we synthesized has gone down for a program. And nowadays, we even try not to focus on speed we try to go after more interesting chemistry and more difficult targets.
Simon Burns: been quite successful in announcing several large partnerships with pharma companies, notably Pfizer and, Sanofi. First, congratulations are in order.
Really, remarkable. Second, there are many computational bio founders who, look to replicate that, success working with pharma. What advice do you have for them on approaching partnering a pharma?
Alex Zhavoronkov: Well, first of all I think that those partnerships with pharma are very important in the very early stages for learning and gaining experience in drug discovery. So, especially that’s very important for computation first AI powered drug discovery companies that prioritize the algorithm and the data over the drug discovery expertise.
So, for example, founders like myself who predominantly specialized AI and computational sciences, and did not have drug discovery experiences before so it took me five years before we really what pharma is.
And maybe 50 to 70 pilots later will realize that you also need to discover and develop yourself and that’s the best way to convince the pharmaceutical companies to work with you. So, I think that my main advice is that if you can to the level where that for a pharmaceutical company to partner with you, come to you, not the other way around.
Because they need to solve the problem, not just run a pilot or evaluate whether the technology works or not. If you have validated beyond any reasonable doubt and you aren’t humans, and you have multiple other programs where you can show the data and demonstrate that you can go after very novel targets, very difficult targets and you can discover those targets and also prosecute them with your own chemistry.
And the chemistry looks novel it works. So, something that would be very impressive to medicinal computational chemists when they look at the resulting programs. So, the level of validation that you need to achieve nowadays is much higher than what we needed five to seven years ago.
So, focus on validation and keep the patient in mind, right? So, I think that you need to be ready to go all the way into humans, and preferably go all the way into phase two, phase three yourself. Because at the end of the day, that’s what we are here for.
So, we are here to deliver the drugs to the patients. And if your ultimate objective is to partner with pharma well, you probably are there for a wrong reason. So, if you are there just to make money or to prove the point or to publish a paper, this is not the game that you need to play, right? I mean, a lot of people do that.
But at the end of the day, our job is to get the drugs to the patients. And if you have this objective in mind and you are very committed to delivering a drug to the patient, you’re likely to be much more successful, you’re going to be much more careful in planning your program.
You’re going to be much more careful when making certain statements. And you’re going to speak with pharma using their own language, because many heads of the therapeutic area would also have this objective in mind.
So, there are agreements also to get the drug into the patient. And when this objective is aligned and they know that they can trust you and you’re validated it’s a marriage in heaven.
Alex Zhavoronkov: So, yeah, I think that one, the objectives both companies are aligned and both of you are very committed to the ultimate outcome where you deliver the drug to the patient. This is much better than when you’re partnering in the mode of a pilot.
And trying to prove the point that your technology works to pharma, right? And very often, those pilots, take a very long time and lead nowhere. So, you might pass with flying colors, but the company will tell you, all right, so here’s your legal payment.
And sorry, we have other strategic objectives. Happened to us times. Well many, times. until we demonstrated that we can commit to the project and go all the way to clinic we couldn’t partner at scale. There are some startup companies that managed to partner on the PowerPoint, so very similar to getting investments, right?
So, [00:15:00] some people are just very successful in doing that and they can get a couple of big names on our board wine and dine the right people at the top. But unless you are committed to get to the clinic with your own asset or with a partner’s asset and you know that you have this capability, I think that you are not going to survive in the long term.
Because yes, you can make a big pharma deal, but if you cannot deliver, and also you cannot deliver on many projects in parallel, you’re going to be dead. And that’s what’s happening right now with many, AI powered startups. So, my advice is that validate, and go yourself as far and demonstrate the pharma with a partner.
Simon Burns: Insilico is now a clinical stage company. Congratulations on moving your first asset into the clinic. In Vial, we think a lot about the pain points in clinical trials. I would imagine just as much as you think about the pain points in, drug discovery receiving technology having an impact in, clinical trials as it would impact the space as much as you’ve impacted-
Alex Zhavoronkov: So, sure. I think that for companies of our type where we also specialize in target discovery it’s much easier to plan the clinical pathway because we discover the targets from ground up and usually it comes with a biomarker.
So, when you utilize AI to discover targets, usually, you actually start with a biomarker. So, you try to figure out which group of patients has the least heterogeneity in terms of target importance you’re looking for the target that is likely to significantly benefit that particular patient’s occupation.
So, if you can clearly utilize this same biomarker that he used for discovery in the clinic you are more likely to succeed. And also, there are many, approaches to derisk your clinical trial and ensure that you deliver maximum benefit to the patient.
So, I think that the most important part when working with a clinical trial CRO is to ensure that they have the capabilities and they have the ability to utilize your biomarker for patient enrollment, for patient selection and also, for monitoring.
And they can provide a really high-quality data and even maybe analyze it themselves in order to to come up with the proper, data path… that will be convincing to the regulators and increase the chest fully and benefiting the max ensuring that you’re enrolling only those patients that are the most likely to respond.
So, I think that’s, where technology meets the clinic. And what we are doing at Insilico, we also have a tool called InClinico, which predicts the outcomes of phase two to phase three transitions. So, they are… will utilize massive amounts of preclinical data and also clinical study design and look at more mainly two scores.
One is target choice score and that’s basically how relevant and implicated the target is in a disease and how heterogeneous is the target in the patient sub population. And the other score is the clinical study design score.
So, one is utilizing multiple AI engines predominantly trained on Omics data. And the other one is a massive transformer deep neural net, which is utilizing predominantly PACs and other data types, but it’s transformer.
And with those two tools we can and we always do, process every one of our internal programs and try to understand how to de risk it from the perspective of phase two to phase three transitions. So, our main objective in the company is to actually pass phase two.
Phase three is more tricky and usually it doesn’t fail as often. But our objective is to ensure that phase two flies with flying colors, right? And to do that, you need to ensure that your target choice is right and you’ve got the biomarker to narrow down the population that is likely to respond.
And also, you need to ensure that you have, of course, the chemistry that world class and is ultimately safe. Actually, for that purpose, where three elite programs, they’re addressing chronic diseases. So to, do that, we need to ensure that the drug is fundamentally safe, right?
Because people will be taking that up in the course of many years. And if you want to have a potentially blockbuster therapeutic you need to ensure that it’s safe and people can take it over long periods So, what we’re looking for in part in terms of mindset, and of course, the capabilities and experience running clinical trials, very specifically these areas. So, that’s always key.
Simon Burns: I’m going go to our last question, which is something you’re very passionate about which is the field of, regenerative medicine and, aging, you’ve made a large commitment to, donate your, life’s earnings to the field. I’m curious to get your, thoughts on the latest approaches using Yamanaka factors.
There’s now several companies in the space, Altos many other. Many are now looking at, small molecule, driven approaches after some recent fantastic work from, some academic labs pointing to an, approach it could be effective. What, are you tracking?
What’s your sense of the, promise of some of these technologies? In general this is something [00:20:00] that, the world is moving towards it seems a focus on a, novel approach to regenerative medicine. Where would you rate at the state of, research? How far from the clinic do you think we might be?
Alex Zhavoronkov: So, sure. Um, we are a big advocate of dual-purpose therapeutics. So, drugs that target a disease in the first place, and also the fundamental mechanisms of aging. So, one of the hallmarks are more than one of the hallmarks of aging.
when we prioritize pro internal programs at Insilico, we of course also are trying to find those therapeutics that have this dual purpose, so we target the age-related disease.
So, for example, our LEAD program is an anti-fibrotic drug that was partly derived utilizing aging research and deep neural networks that are trained to predict human age, human biological age but also are very implicated in multiple fibrotic processes.
So, we’re looking for master regulators of fibrosis and the targets that are implicated in aging. And the first program is already in phase one human clinical trials for lung fibrosis. But we also achieved preclinical candidate stage in kidney fibrosis for the same target and are almost about to achieve preclinical candidate in skin fibrosis.
So, it worked in pretty much every assay we tested them. And my bet is, again, on those dual-purpose therapeutics. We publish the research paper recently where we highlighted over 140 targets that are likely to have dual purpose structured by done novelty confidence and drug ability, and commercial tractability.
So, you can take this paper on dual purpose therapeutics and form a bunch of companies from, this paper. We also published multiple other papers of ALS, in certain cancers, where targets have dual purpose and we’ve validated all the way to animal models out in the open.
I think that the stage of longevity for pharmacology, the state of longevity pharmacology is pretty advanced, but we’re still in the early days. So this, industry is definitely uh, going to explode. Because this is the best way to discover potentially blockbuster therapeutics that maximize benefit for the patient.
Again, in our case our fundamental value in the company is patient first, so we want to ensure that we maximize patient benefit. And if you are targeting aging disease at the same time, even if the target is not number one most important target in a specific disease usually it’s not same target is… if you are really hitting the number one most important target.
So, many diseases, it will be exactly that, right? So, you are going to to max… patient benefit due to an off-target effect that is also targeting aging. My thoughts on reprogramming, I think that it’s very promising technology but it’s very early days.
And there are many, reason to believe that it’s not going to work. So, we should wait and see. It most likely will work in certain diseases.
For example, many ocular diseases ocular diseases are likely to be very treatable with reprogramming, multiple diseases of the skin and diseases that were reprogramming would probably be to a specific tissue preferably the tissue that you can also easily access and monitor.
But on a systemic level, I don’t believe complete or partial reprogramming is going to work primarily, because there are many sources of damage that accumulate during aging. And some of those sources of damage are not controllable using pharmacological means.
Like for example, if you open up a 70-year-old on the operating table and feel the aorta it will be much less elastic than the young person’s aorta. And if you let it dry, it will be like an egg because the tissue has mineralized, it’s calcified.
And there are many advanced glycation end products. So, for stem cells to move in, it’s very difficult so any kind of tissue recycling constrained. For the cells to find the right niche it’s also very difficult there is a lot of fibrotic tissue So, if you think fibrotic are higher than this ultra-calcified and the extracellular matrix is already very stiff. You’re probably mistaken, right? So, it’s not going to give you a huge benefit. So, we need to look at maybe organ replacement as the next frontier in, regenerative medicine.
And I think that would probably give you the maximum longevity dividend. But pharmacological approaches, I would focus on dual purpose therapeutics, regardless of whether they are focused on reprogramming.
Or, if it’s analytics, if trying to target the senescence cells, those old cells that just sit there and do nothing and [00:25:00] excrete toxins in the microenvironment.
Or if you are addressing fibrosis, again, that’s my favorite. Because I think that you can really gain maximum benefit if you’re addressing fibrosis as a disease of the biological process driving multiple diseases. Also, if you look at stem cell exhaustion telomere attrition epigenetic drift and many, others, right?
So, there are many areas that we can go after. I think that we need to demonstrate that low-hanging fruit first and showed that you can achieve a blockbuster therapeutic by targeting aging and disease at the same time.
And afterwards, once you have the resources to move forward, you can explore other methods. A young bullish on reprogramming, but also very conservative and cautious. I think that one of our really cool tools that we are developing is a fully robotic AI driven laboratory that is able to capture a lot of data from the sample.
So, the sample it goes in, I’m not going to talk about which sample is going to be, we will be able to work with multiple different sample types. But we’ll get deep phenotyping data from the sample and afterward, the transcription lesion response and [inaudible molecule late pretty substantial So, those data types will be able to actually even help us identify those reprogramming candidates more efficient. write and design some of them from scratch. But that’s not the primary goal, right? So, we, of course want to look at the targets that are coming from epigenetic data and combination of epigenetic and transcriptomic data.
Simon Burns: Fantastic. Well, with that Alex, thank you for the time and appreciate you taking the time of your day for this conversation.
Alex Zhavoronkov: Thanks very much for having me.
