My guest today may be one of the smartest men that I’ve ever had the pleasure of speaking with Alex. Chevron Coff is currently the CEO of in silico medicine, a company with the goal of using artificial intelligence for drug and biomarker discovery. Getting a drug all the way through clinical trials is an extremely expensive and risky process, typically costing more than $2 billion with an over 90% failure rate. Alex wants to use the technology at his company to drastically reduce this failure rate, saving time, money, and ultimately human lives. In this interview, Alex educates me on exactly how his company is going to do that, how the technology works, and just how key the role of artificial intelligence will be in the longevity and biotech industry going forward. It’s a fascinating interview that unites the world of computer science with the world of biology. Without further ado, Alec, Chevron cuff. Alex, thank you so much for joining today.
Yeah, I was when you told me that we had to have this interview on Saturday. Yeah. I jumped at the chance any opportunity to talk to as someone obviously and as intelligent as involved in the longevity industry as you are. I want to start off by just talking about who you are, your background, because I did quite a bit of research on you and you started off in computer science, correct? Correct. How did you get into longevity and that’s obviously a pretty big jump to go from computer science to longevity. You’re using your skills in computer science. I’m curious what that whole process was like. And I’m also curious, since you’ve been in the industry for a while now, how are your biology and biotech chops squaring up these days?
For sure. I’ve been interested in longevity for my entire life since I was a kid. So I don’t think that, uh, aging is such an appealing process in general because you work, uh, for your entire life towards some goals that you’re saying are important. But at the end of the day, you start losing at some point in time. It’s not possible to step on the path of continuous improvement. So after 44, a male, I’m actually earlier for female, many systems start to decline for me, I’m not contempt with that. However, I was born, I know for the very special point in history when the Soviet union collapsed, I was born in Latvia and at that time there were not that many opportunities to pursue a medical or a scientific career because those professionals immediately lost an appeal because it wasn’t possible to make any money there, right from, of course.
So as fields saw how lost more than others. And that’s why also I’m so afraid of the next economic crisis because people, I live for this one. So when I saw that, um, people start prioritizing immediate gains and the immediate issues in favor of longterm goals. So I decided to go into computer science and also commerce once I graduated from high school. So I moved to Canada, went by myself and um, started, uh, my studies of uh, Queens university in Canada, in Kingston. So I did two bachelor degrees, computer science and commerce and commerce, also with a focus on finance and mission systems and the afterward, uh, to work for a number of semiconductor companies. One was EMC Sierra, which is computer networking. And actually that experience gave me a lot of thought and a lot of interesting ideas on how to work with signaling pathways, instrument biology as well because of course, uh, neural networks are somewhat resemble some information flow in a human body.
Then I worked for ATI technologies that was at that time, the second largest or largest designer and the producer or graphics processing units, GP use those such ships, those semiconductors are now powering our show intelligence systems, uh, worldwide. So, and media is one of the core providers of GPU for a machine learning. And neural networks are now kind of powered almost entirely by age abuse. Of course, there are other technologies, but to be used as a core one and in 2006, AGI was acquired by AMD. So even before that I started my graduate work at Johns Hopkins doing my masters by technology with a focus on by medics, just to start getting into biotechnology deeper after age, I go to choir, they had some more stone options, so they them, but I managed to make some money so I didn’t need to work in industry and decided to pursue my dream to go into biotech and specifically into aging.
So I completed my, uh, greatest studies of Hopkins and worked for a number of biotech companies and the usually people in aging. And you’ll notice that a lot who are more forward thinking more. Some of them come from more HD backgrounds or from computer science, social already gray. For example, he a computer scientist, he is a computer scientist by training and there are many other people who work on computer science, on the information technology. They lived through several revolutions during the revolution in personal computing. We saw that something that was previously impossible became possible. And then, uh, mobile devices and the internet and the social networking. So you see the technologies on technologies become exponential and they become daily life. Things that we previously thought were unimaginable, become imaginable and become real naturally read and wait for a flight. Once you live through a few of those, you’ll naturally started asking yourself, so what’s next?
Why don’t we take something like aging on their biomedical control and white envelope? We don’t reprogram our bodies or build some kind of roadmap to prevent disease and eventually solve aging. So I decided to kind of take a little bit different routes than most other people take in this case. And I actually liked the way Aubrey degrade. Did it. So she, instead of investing into those fields and self right away, he decided to get educated on the subject, commit himself religiously to this subject and himself became experts in aging. And at the same time she managed to involve a lot of people into this field. So I found that probably in my perspective, Aubrey is probably the most impactful. I’m curious, scientists are employed by drunk apologists in the world and he is no doubt the best example polo. And of course very early in the days he had to face a lot of controversy and a lot of public criticism once he just started.
And nowadays it’s much more accepted by this kind of, I’m not going to make an exorbitant statements with my head down and started just working in academia. After my grad work with Hopkins, various studies, I decided to explore multiple areas of biotech cause you know, aging especially a multifactorial process. And there are so many variables and so many different levels that are involved human biology and psychology in the process. So it’s very difficult to understand what’s driving it. And it’s not one single process. So it’s trivial, so immediately want you to just join us field with little prior knowledge. You’re saying that there are many low hanging fruits and on the realize the complexity of this process on you understand that it’s important to dig deep into individual fields. So I even worked for a company called GDC bio at that time. It was one of the largest conference organizer by fortune conference organizers in my space and I decided to work for them for a couple years and the, to know multiple fields of biomedical research, so organized more than 30 biotechnology conferences in the first one and a half year is the STEM cells in infectious diseases in LA Regina.
It’s not as some drug discovery, drug development. So I quickly grew a very substantial network of scientists and entrepreneurs worldwide. So dozens of people. Also I got to know, and of course when you’re organizing trans youth from this position, you have the ability to really interface or interview a lot of really brilliant scientists and that’s what they did. So you essentially become part of the organizing committee and the scientific advisory board all the conference to get to know all the thought leaders and it gives you a very broad perspective so you understand where each field is. And that my map, a Charles counter, the former director of the human genome project, so there were three James walks on Frances call on us now, the director of the NIH and Charles counters, very prominent geneticist. I’m a bedtime, she was at Sequenome so I started consulting for Sequenome at the time I moved to restaurant at some point in time they want to set something up there.
And I’m also the founder of my PhD in physics and mathematics, muscle States university, also in rationalization of amino acids in proteins in the course of aging. Some other big topics to discuss and open a lot, but the very large pediatric cancer hospitals over there. I was a foreign scientist and I was managing kind of the bioinformatics and regenerative medicine lab. Very new, very brand new source, super top notch children’s hospital where they do a lot of bone marrow transplantations. So I wanted to understand that aspect and of course blood transfusions because that’s one of the areas where you would be surprised how high the survival rate is. So those kids, if they’re treated properly, they survive more than lens sounds all the time. And the procedure is involved in kind of getting them better are in a way similar to what a rejuvenation therapy is.
But it looked like it was balling mile at transplants with black trans blood transfusions, a cell therapy in some cases. Now people explore gene therapy. I need to do the sequencing. You have to understand the genome to understand the underlying that much stuff occurred to understand how to treat the disease. So at that time I got a lot of experience in this area and uh, the moment some of the early algorithms for grouping genes into networks and understanding how those networks change on their state over time. Specifical longitudinally, um, from norm to disease and uh, and uh, completely transferred to bioinformatics. I went to Baltimore again, set up the company and that’s how it Selecto started. So to answer your question, that’s my kind of biographical note of, to answer the question, how do I transfer? How would one transfer from IC to biotech? It’s jumping into the opportunity.
So that’s was my biography. That’s how I switched from a information technology to biotech. And again, my advice to everybody who is swinging a switch. So it’s very important to really go and do your work from both academic side and uh, maybe spend some time in the industry because unlike any other field where very often you can jump into without any formal education. So biotech is not like that and it’s not it where they can hack your way into the system without having any limit credentials. In biotech, that’s very important both for reputation but also for yourself because you need to really dig deep into biology works, understand the central dogma, understand. But all in your degree about genetics, cell biology and many other aspects from biotech need to understand the how drug discovery processes work. How long does it takes you to discover a drug and put it on the market?
And why is it taking so long for why are we so inefficient? Until you really dig deep into those areas, things are going to look much more rosy. And a more optimistic is just biology is immensely more complex than any other field in human kind of on your radar. And the chemistry, which is extremely important also to be able to develop drugs is also extremely important and it’s also very complex. So bridging biology and chemistry is very difficult and they need to do the work. So after I switched from mom biotech, so I was starting to work for my chief for biotech, I decided that one of the problems in li and having problems in the industry is the inability to properly track the funding, going to buy a tank. Because currently there are lots and lots of funding opportunities for scientists from the government, from the industry.
And the NIH spends about five plus billion dollars annually. So the U S is by far the largest funder of science, biomedical sciences. And I’m just wondering since you didn’t learn in the U S is spending 45 billion, there are additional sources of funding like national science foundation, like 5.5 RG actually you in the army spends quite a bit on biomedical research and then there is European commission, there is Australia, Canada, many other national governments who are spending billions and billions annually on biomedical research and understanding where this money is going and how is it spent and where does it result in scientific publications or research data or other malicious of sources that you can tap into later on is a, as you know for known to understand what’s happening on a public sector before you start investing anything yourself. And we built, I again joined forces for multiple scientists and we built a system called aging for Tulio orangy or the international aging research portfolio that tracks our funding from all over the world and classify as this funneling into buckets and on trays as the days with time to understand how scientists pros form those dollars into research papers and the patterns of the clinical trials and into the products from the market.
And also into the pockets on beta, like genomics data, transcriptomics data, proteomics data. So what we call omics and also chemistry datasets. So high strip was being experiments, et cetera, and we started tracking that time that was 2011 about a trillion dollars worth of funding. Now we track about 1.71 point 9 trillion. If you look at this funding, very often scientists who bank, they are in aging research, they actually do not contribute as much to life extension for example. So somebody might be studying the diet. So for long lived organisms, some long list humans or somebody is going to be studying aging in a workplace or preventing falls in the elderly, et cetera. So they are not going to be contributing as much to the growing body of knowledge that will lead to a significant plant extension, right? Because we don’t see anyone on the bias.
So living in the past hundred 20 and there’s been thousands with generations of humans who tried all kinds of foods, tried all kinds of diets. That was somehow the oldest living human on record will live to 122 and a half. Or if it’s true, she smokes. So when I was seven, she drank wine up until the end of her life. And there are multiple other super centenarians who do not really have any specific bias or exercise teams. They just, uh, are genetically predisposed to longevity and they are just lucky. So I think that many scientists who clay in the bill do aging research, they actually do not try and just demographers or statisticians and the contribution of their work while making big headlines is run or marginal. However, many other scientists who are studying, for example, counter or neurobiology or a specific brought the user specific pathways might have, might be making substantially bigger contributions to science than those demographers and make things substantially more contributions to aging research than demographer is by studying cancer or Alzheimer’s or Parkinson’s.
And they will not even classify themselves as aging researchers. They would say, Oh, I’m doing cancer research, or I’m studying regeneration. A really brilliant female scientists. So scientists, um, scripts are, she cloned mice from what induced pluripotent STEM cells by essentially injecting my cells, reprogram myself inside of most embryos and cloning perfect copies, all four born or mice by a young blastocyst injections off cells into mice plus the system that implants and goes bust, assists signpost mice. And I’m sure you does not consider herself an aging researcher. And I thought, wow, those are breakthrough experiments that will lead to those experiments in 2018 2019 time 2008 2009 so some years ago, that was when I’m a Yamanaka factors was just discover. It’s so brave for the work and published in top journals and she never considered herself to be an aging researcher, but I would glance and find her aside.
Our efforts on the analyzing grounds were also classifying people by their contributions to specific research areas and also by the contribution to aging research. So I basically started looking at who is, who in the area and constructing those pathways and graphs of factually people and looking at their enough contribution to this field. So to start in design my own career, but there’s somehow to make a bigger impact. And I also decided that the biggest impact I can make in aging is to actually not even call myself a Badger and biologists have go into bioinformatics and icy. So back to her, I started to analyze massive amounts of data that’s being generated and put this data to work or aging with search. So first by analyzing text, then we started aggregating Norma someones so far Omix data from public sources and looking at how the various beta types but very various States change during human life frame, you know, young to old and from all to very old to understand what kind of processes are um, Kering and how can we intervene by big data analysis. And that’s how I started on silicone. So I think it’s very important to understand what are the most impactful fields in this area. Who is who and then understand where you want to go yourself to make the biggest impact. So in my case, it was actually going back to it and doing some very basic work in bioinformatics and signaling pathway analysis. And this is about what you aging systems, that was your counselor looking at multiple other diseases that are age related. And then I switched to chemistry. So we realized that in order for you to manipulate biology by, for example, modulating the expression of specific targets or specific proteins that are responsible for some biological processes, you need to have the ability to very quickly generate really with chemical matter that would inhibit or activate a certain biological targets.
And for that we had to do a few breakthroughs in chemistry. So we developed a system which allows you to, with atomic precision design molecules that hit specific targets to be able to validate those targets quickly and to accelerate the pharmaceutical drug discovery drug development substantially. So essentially by focusing on aging and to understand how to be more effective in this field, we managed to develop two fields, foreigners generative biology. So working with biological data, using AI and generative chemistry, substantially accelerating discovery by accelerating from design and making important more efficient than the high quality. Using generative chemistry approach is using AI. So that’s where we are and we are making money. So in both attracting funding from break credible investors, even though we are also advancing the field of aging research.
Alex, I got to say it’s a very impressive resume you have behind you and all the stuff, all of the little pies that you have your hands in. It’s so much more than the average person. And I just got to say it’s really impressive and you touched on a lot of different things here and I kind of want to go back and touch on some stuff in particular. You talked about the impact that aging researchers are going to have on things and in longevity as a whole and how it’s not that big of a deal. And I think that’s a pretty important thing that I’ve realized recently. I think a lot of people are starting to come around to the kind of habits that you hear about like fasting and staying out of the sun. All those things aren’t really going to be a huge breakthrough.
It’s really going to come. What’s really gonna push the needle to where we see massive changes in aging will come from the molecules that companies are able to discover and that companies like you will be able to help discover. So let’s talk about exactly what in silico does. You guys are working on drug discovery, working on finding biomarkers, but, and you’re using AI to do all of this, but how does exactly, does that work from a kind of explain it like I’m five perspective cause it’s pretty hard for me to wrap my mind around how an AI can figure out a drug that is going to work with the human body. Before I do that,
I’ll ask you to quickly do a thought experiment. It will be very easy to translate from something that you know very well to something that you might not know very well. So from a data type that we are very comfortable with the data types that’s yours might not be as comfortable with. All of us are very comfortable with imaging data, right? So we are all good at recognizing fixtures, right? So thing called for right now yourself as a deep neural network, as a form of AI that is exposed to a data type that last few months are very good at as well. So like imaging data. And now imagine that I’m showing you a picture of a person and you’re looking at somebody who has gray hair, who has all the characteristics of a male and um, has losses. So what would you say the age of the person is? So by just this description is actually a question to you. For example, James, what would be your guess? How old person is
likely above 50 or 60 years old? I have to see the picture to tell, but yeah, with gray hair,
what other features will you see on the person’s face in addition to just the three features I described,
wrinkles, not saggy eyes, but definitely wrinkles in the eyes. The crow’s feet, definitely skin will be a big factor in that.
Great. So, and how would I make this person look younger to you? So if I wanted to make that person younger, basically decrease the perceived age of the person, what would you do?
Definitely start off with the wrinkles. Smooth out the skin, turn the hair color to natural. No, no gray. That’s where I’d start off.
Great. So we just, uh, rejuvenated the person on the picture, right? Using a very simple phone experiment. And now let’s kind of explain how it would work on other days. Think about it then. Fixtures again. So deep neural networks was proposed a long time ago and there’s been more. So for fundamental studies or form, the deep learning in the M and a previous straight. But the real advances in deep learning started, uh, after 2013, 2014 where you can, your own networks outperform humans in image recognition, text recognition, voice recognition, so many other fields. And uh, well, uh, the funeral network started outperforming humans in image recognition and with, you know, recognizing cats, for example, images just by looking at the ear, sticking out, thoughts behind from behind the college, for example. Those tricks, uh, are very impressive, right? So you see an AI system annotating a picture better than a human. So one of the areas where deep neural networks I became also very good at is recognizing the person’s age.
So I would just like for how you describe, uh, predicted the age of a person just by from two features, uh, where you could actually predicted from one right for your hair. Do you feel networks became quite good at predicting the class? For example, you know, the age group, so young, middle aged, old and not long getting to basically buy months precision, right? So now if you are using a picture, the deep neural network can predict your age within, you know, two years, so better than many other data types. And now we can also train those deep neural networks to interpret this neural left for us to derive. The most important feature is give out your age and you can lead. Surely the doc, what features do you want to correct to look younger? And now in addition to pictures, we can include other data types, right?
So you can use gene expression data. So it’s essentially they’re not totality off your genes that are read at any given point in time, the car transcribed. So for those of you who don’t know what gene expression is or protein expression is, just Google it, it’s not worth while, uh, you know, learning more about aging if you don’t understand those fundamental science. So it’s very important to understand the central book moral biology and basics of human development. So you can know a lot for a schema. Now predict your age using gene expression data. So the number of genes that are transcribed from DNA into RNA and not to tell it too far and they made it the they use. So you can use proteins as a protein, as your data source and on your train deep neural networks on millions of those datasets of gene expression, protein expression, annotated with age to start guessing the age of the patient from this data correctly.
And you can play around with this data. You can, for example, group those genes and the signaling pathways. So reducing the dimensionality of beta and training the [inaudible] pathway scores, a young predict age. And you try and do this in um, using the data coming from more or less healthy people from longitudinal studies where a lot of healthy people were profiled predicting age using this data type. Once you train a predictor to guess the person’s age with very high accuracy. So by far better accuracy than humans. And of course humans can not even guess the age of another human. For example, just by looking at gene expression data, that’s where you’re looking at a table off [inaudible] with the numbers of basically tell you how much of this protein is being expressed. So the quantity of this protein in the cell he wants can both work with this data but pushing a scam and you can build those very accurate predictors and then make them an explainable.
So you are trying to look at how those quote unquote molecular wrinkles start before and since we now understand how to return the expression of certain genes with small molecules, with drugs or with other interventions like temperature, dietary restriction, sleep, etc. We can not modulate the expression of those genes and start reducing those wrinkles, quote unquote in the molecular data and trying to make you younger so that you can grow lab work by administrating certain interventions, either chemical or behavioral or gene therapy. Many, many other kinds of approaches can be used to modulate a prediction of a and see if, uh, this kind of intervention would make your younger or older make the patient younger, older, or just like with pictures for example, reducing the wrinkles might not necessarily make you live longer. So after identifying those promising kind of targets for interventions, and after developing those interventions, you also need to prove that those interventions are extending human life and reducing mortality.
And that’s where you start looking whether the people who took specific drugs that are modulating those genes and proteins that make you look younger when perturbed. So the deep neural network, you need to look up, uh, whether in the past somebody had taken this drug, started living longer or uh, became less sick or a preventative counselor in the large population. And you start proving your hypothesis using retrospective elevation, looking up large population, old databases where a lot of people are taking different drugs. And uh, we did that as well, whether the silicone. So we basically showed how to correct those molecular wrinkles. And then you’re also young when you’re working with pictures, for example, very often a good doctor would be able to look at the person and say, Oh, you are sick, right? Oh and you look older than your chronological age. Maybe you know, you have this disease.
Very often a very good doctor can see whether you are sick from, from a picture. And actually for example, think about right now, imagine the a kid with the balance disease or with like celebrated aging progeria. So if you ever saw one or two or three, you can now imagine them quite well and you can also recognize some quite well. So now the thing called a thought experiment where we’ll do this with another datatype. Again, gene expression and protein expression data. So we will train our deep neural networks to recognize age very well and then we can retrain them on specific diseases. So we can now retrain them to recognize specific disease like for example, for G area or fibrosis or a specific type of cancer. By retraining those deep neural networks, trained on millions of samples on just a few hundred or thousand patient samples with disease.
And as soon as we before understood what are those kind of molecular wrinkles that change over time, we can now see what this changing between more model aging, quote unquote Al disease. So now you can see, okay, well what makes this patient a look like a patient with a specific disease as opposed to adjust as a healthy normal person for that specific age. And we built algorithms that uh, basically learn on millions of examples, annotated with age healthy people and then, uh, retrain on very small data sets of uh, patients, silo, patient beta with specific disease. And then looking at what kind of feature is, what kind of factors can be influenced with drugs in order to return back to the normal healthy state or completely terminate the life of a cell if it’s counselors or if it’s driving the rest of the body in the wrong direction.
And I’m using those thought experiments and pictures. We started working with many data types, not just with gene expression data or protein expression data. We started working with methylation beta. We started working with microbiome data. We started working with blood biochemistry data and started also integrating those data types together because a young one unifying feature that basically can unite all of the data types together is age. So everybody has age. So if you know how deep neural networks look like, you can basically imagine that going into the deep neural network you have many, many, many features, uh, thousands of them, millions of them and at the end or that you can own that for the output layer, you will have just one you wrong with predict age of the person. So you can integrate many, many different data types that previously would not be even connected.
Like for example, you can connect all pictures with a voice and pictures with a gene expression data and once you train the deep neural network or multiple data types to predict age, they actually learn biology and they start forming nonlinear associations between the different data types that were previously completely disconnected. So at those silico we basically unite aging with search with a really advanced understanding of human biology systems biology. But the sound was different between aging and disease and that’s how we hung for molecular targets or protein targets that are driving disease and we can look at multiple diseases and then we have four. The ability to very quickly generate novel molecular structures of shift those targets and essentially correct the disease space back to normal. Once we, when we think about aging as a disease, we can also create a lot of experiments where we are probing specific protein targets to make patients look younger to the deep neural network.
All starts with a thought experiment. When you are thinking about any beta five as a picture, you have to cite that and you have to know that the deep neural network will always outperform a human when describing this picture and recognizing this picture. And I’m also interpreting this picture and also imagining certain pictures with specific properties because you can hold on for X number of forget on one day, learn from millions, they outperformed humans and then you can use the uh, assault experiments to extrapolate on other data types that might not be as obvious and as common for the human mind as pictures.
Okay. So if I’m understanding correctly using the thought experiment, that picture, you know, um, basically what these neural networks can do, and this is obviously a very greatly dumbed down version of everything you just said, but using the picture as an example, what these neural networks can do based of biomarkers of aging, biomarkers of disease, that’s kind of like the target, I guess. You have all these data sets from millions of studies, a hundred thousand studies, a hundred thousand pieces of pieces of data where the neural network pulls all that data. And it’s doing these, I guess, experiments inside of the neural network to essentially unwrinkled the picture to fix these biomarkers of aging biomarkers of disease. And is that somewhat accurate?
Oh yeah. You’re kind of saying enough. It’s basically, so just we don’t use the picture is when you use other data types that are more biologically relevant. The thought experiment is performed using a picture.
Yeah. And how accurate, well, two questions actually. How are you testing these results and how accurate are they? Obviously if you, if the AI produces a molecule that you think is going to work, it’s not really something that you can test very easily. I mean with regulations and all of that. So I’m curious how accurate it is and how you’re testing things.
I must say that. So that’s where you really need to spend some time on in the academia and the industry to understand how complex those things are and how difficult they are to validate Paul. How often do they fail and for what reason? Even with a traditional approach for drug discovery, usually the process of identifying or normal protein targets and developing a molecule for it and taking it through the clinical studies, it’s usually, it takes about a decade. So 10 years and costs about $3 billion with different phases. So costing sometimes up to half a billion, a billion dollars. And um, in our case we are very often in that territory where when you have to validate a novel target with normal chemistry, even the whole experimental system. So everything that you are doing is new and the failure rate for those experiments is usually very high.
So again, on a traditional drug discovery paradigm, if you are discovering a drug available at 92% of the time, so unlike many other industries where you have a lot of successes to train on here you have mostly failures, train them at a silicone. We started first doing target validation for this time, whether we take the right driver is older disease and with the many, many experiments, so ourselves and also in collaboration to validate our target identification algorithms. And they usually do that when you’re working with patient beta or when you’re collaborating with somebody like a pharmaceutical company or by full legit company. And you’re doing this for somebody else already has the very good targets validation mechanisms. So they already have the ability to very match the targets for the molecule and the molecule in the experimental system that they trust and that might be relevant to the disease. So we spent about two years validating our target discovery capabilities and we managed to do that on multiple diseases. Primarily. Counselor unfortunately only can be confident that your target discovery system works after a phase two clinical study in a human where you showed them the drug is effective in the large human population and you’re really hitting the right target prediction to be effective. So we validated an experimental model. We failed a lot the very beginning ourself. Uh, and of course also with academic research partners cause uh, usually you start collaborating with many, many academic labs where you can validate. And then, um, once we validated the target discovery system, we decided to validate our chemistry system. So where are we can generate, uh, with again, atomic Racine precision. We can imagine for specific targets. And there we start at those. Basically you think of fell on last launching space X walkouts with just some mobile technologies in there. And they were blowing up on the launchpad, right? He had no few, few launches left before bankruptcy. So we have, we were in that situation very often with molecules. So when you are validating the molecule that is generated completely with AI, you need to invest in synthesis, right? So you have to make this molecule, you need to build it. They cannot just buy it from a library, which is also expensive. But synthesis is a very often than the order of magnitude more expensive. So you make this molecule from scratch, you order an experimental system to pass this molecule, and if it doesn’t work, it means that you spend sometimes a year or half a year and few hundred thousand dollars or sometimes millions and fail. So we were a death in that state very often. The thousand 17, we started validating those molecular generators. So we saw that even conceptually, they work, they didn’t produce the best molecules out there, not that the level of gray medicinal chemists, but uh, already be, we, we saw that they are working in the right direction.
So you’re getting some experimental evidence. And we did a back van with Jack kinases with DDR kinases with simpler kind of Argus that are also easy for human medicinal chemist to work with, but also understand how our AI did this. And now in 2019, we started, uh, we validated those systems for very complex started. So previously it wasn’t possible for human medicinal design molecules for complex proteins targets quickly. And we’ve managed to do it completely with AI and a driverless mode. So without, uh, any human intervention. And yeah, so now we are trying to combine our target discovery capabilities with our chemistry capabilities and basically best as many are new molecules targeting specific proteins that are implicated in a specific diseases like fibrosis or for example, that’s one of my favorite, uh, biological processes of I study and also senescence and the, um, various, uh, cancers as well.
So where the biology is clear, liberate, perturbed and we literally this regulated and where we come, uh, by manipulating certain targets of the amount of our engine where we can modify the disease and there is a clear ass site to test it. We’re already validated this some fibrosis. We are of course looking at other diseases so currently I think my system is very well validated in both biology and chemistry and of course we are performing more and more validation but right now there is already at least three programs. Internally we are, we have all of the targets and the molecule coming out of our engine.
Yeah, that’s super exciting stuff and I think one of the cool things about AI is the the fact that things kind of grow exponentially as it gets better, it’s just going to get better and better and better. Which kind of brings me to my next question is do you think there ever be a point where the AI is so good at figuring out targets, figuring out drug discovery, even regulations or ease for these types of molecules that are discovered where you can push them through? Because it was discovered by the deep learning, the AI that you guys have, it’s able to get pushed through phase one, phase two trials much quicker.
That’s summer egg with DRI, but unfortunately you can only accelerate so much using AI. So even if you are perfect, if identify targets and I’m generating the old chemistry, you still need to validate those molecules and the experimental system to show that they’re safe and to show that they are effective and show how they work with other drugs and to show what kind of side effects they might have. So those regulations are there for a reason. And again, you understand that only after getting deep into this field. And one good example for this is Corona virus for example, the recent or on the virus. So you know that humans companies are very good at making vaccines for most the infectious diseases foremost virus as there is a vaccine right there is non for AIDS or for several other viruses, but many viruses do have a vaccine.
So we know how to do it. Unfortunately the process is very long and tedious. Even though conceptually it might not sound difficult, but the reason as long and tedious as, because you really need to do the work to understand that the vaccine works another safe and you also need to have the ability to mass produce it. So right now, if somebody had a perfect vaccine, they still wouldn’t be able to put it on the market immediately because you still need to validate. And even with loosen regulations, nobody has done that today. So you still need to have, you know, a year, year and a half, two years at the very minimum to test a vaccine for the regulator is they realize how complex it is to design molecules using AI and [inaudible]. So they are already working on loosening the regulations and uh, making it easier for companies utilizing AI to get through to the clinic.
But that change in regulation would result only in very marginal difference. So from the standard year of traditional drug discovery timeline, you might call two or three and increase the success rate of those kind of bets on specific targets cause drug discovery based, kind of like molecular casino saw. You’re making bets on specific targets, you’re, you can make a bigger batch of smaller batch amount you have to learn and doesn’t always land on the target that you predicted and sometimes you might cause some damage. I think that there are rules for drug development. Even if those rules change, you are allowed to put AI discovered molecules into the clinic faster. The differences not going to be Panax is going to be maybe a X difference is going to be in the kids rate. You may actually see off the system. It’s not going to be in the time.
Yeah, that’s what I was going to say is that with two X is still pretty, pretty good if that were to ever get to that point. But yeah, the real benefits of the AI is that instead of having, I think you said a 92% failure rate, instead of having a 92% failure rate, ideally it’s going to go up a lot. It’s going to be a much lower failure rate. So I want to talk about a little bit about Silico the actual company and because obviously there’s a few different companies that are doing AI. When in biotech, I want to know what makes in Silico different, what’s guys’ competitive advantage?
So our competitive advantage is so, so we have several sort of, the main I guess is we were in the right place and the right time in 2014, 2015, when many of those technologies in AI emerged and we started buying uranium. So we basically, you know, before, uh, that was our flow became ubiquitous before many of those AI tools became kind of out of the box. We were already there experimenting and we basically went through many, many technologies when you failures, when you successes and with patented, uh, one of those methods. And also we published some of those methods. So we established some reputation as, uh, an academically focused, uh, AI company. So for most of the claims that I’m making, I have a paper to show where I have a proof of concept at least showing as possible. And we, to my knowledge, were the only company currently spanning multiple of phases of pharmaceutical drug discovery, drug development.
So we can do our get discovery, we can do small molecule chemistry generation and we can also do predictions of clinical trials, outcomes and all that with long AI system. So it’s basically like, you know, playing three different, totally, completely different games for different data types and using AI. And for us it’s one seamlessly integrated platform where we identified a target, a discover, a molecule, build an actuarial models. So the risk model for clinical trials, phase one to face the transition phase two, phase three, transition pylon through another AI company that cannot at the same time discover a youth targets in a variety of diseases. Then for those targets, generate the will molecules and for those molecules on the targets come up with an actuarial model for clinical trials outcomes. So predicting phase one, transition phase three, transition in a variety of diseases. So in that aspect, I think we’re quite unique because it’s pretty much like playing several computer games that are completely different. So super Mario and Assassin’s creed on the mortal Kombat with one system. So in our case, I don’t know another AI company that has the exact same capabilities, but at the same time we also have a very experienced team because very early in our story we started focusing on services for the pharmaceutical companies. And very often those were early stage technology pilots where we had to solve very difficult or impossible problems. And we looked to work with some of the best people, best humans in the world, best scientists because pharmaceutical drug discovery, drug development, those scientists are not on an idea than other universities. They are in the pharma companies and they usually don’t publish as much. They are very difficult to approach every secret of. So unless you work with these companies in very difficult areas, solving some of their very difficult problems, so they’re struggling with it can not really be the same level or above. So you really need to grapple with the best champions in Brooklyn. You need to work on the floor with the best, you know, boxer, if you want to be good in boxing. And we got to interface some of the best humans out there in the pharmaceutical companies.
And very often we impressed them. So my team is not very experienced and they are very committed. They are very committed to both longevity research and also to finding really good drugs for patients, unmet medical needs. We’re also geographically distributed so I can work in multiple geographies and in multiple time zones. So for us the work has 24 seven never stops. We are growing and we also have the best investors out there. So I think the quality of the investor is more important for you often than many other factors. And we’ve got the best investors in biotech and best investors in AI. So some of the really top experts who invested in many, many biotechnology companies. So they saw everything and the investors who are responsible for my multiple, you know, AI unicorns. So they also, the technology, the villages, but they also are helping a lot when it comes to scale up when it comes to some very difficult decisions we need to make all the time. And with introductions and of course it’s reputation.
Yeah, that’s an amazing answer Alex. And it sounds like you guys really have it figured out and so it goes really on top of a lot of different stuff. It’s really exciting. I want to ask a little bit about and silicone business model. You mentioned that you had worked with companies, different companies to help them with drug discovery and all of that, and I’m assuming you have something set up where you could get royalties if any of them get through. I know you mentioned in another interview that I listened to that you guys also offer have nutraceuticals. Is that something you’re going to be going into more as well, producing supplements, etc.?
We experimented with many models, uh, during our, um, corporate development and sometimes people perceive us for what we and have attempted to do, tried and decided that it’s not the great business model. So we have a hybrid model where we develop software, just started selling the software. We also do service collaborations. So people hire us to solve a difficult problem and we’ve done participate in the, so we either get paid, they’ll try or we participate in the kind of future appreciation of the project where we solve the problem. And most importantly, we have our own law discovery pipeline. So we actually have for our own molecules that are kind of sailing through time and accumulating more meat on the ball and so to speak by multiple validation cycles in experimental models in mice. And we hope to go into humans hopefully next year at least those some disease areas like fibrosis, that is the main business model because with services or software, you can generate only so much revenue so you can sustain, but it’s very difficult to come exponential with your own drug discovery pipeline. Some of those drugs can be license for hundreds of millions of dollars, even at the preclinical stage, if you can demonstrate or this definitely normal and works and demonstrate a lot of preclinical data. But of course for us to go clinical boast payouts become an order of magnitude bigger. So signing those deals, licensing deals may be in the billions, right? So we have a few of those switch rates in our pipeline as well.
That’s amazing. And I wanted to ask about the nutraceuticals. Is that something you guys are still doing
Something that we wouldn’t do for a company that is not a big pharma company. So for example, if a big pharma company comes to me and says, look, I’m more working on less nutraceutical, but I want to try a natural products in this niche markets, none. I would do that. I would jump on the opportunity, but we don’t make ourselves a number. I’m going to tack it on. For example, we come up with research for somebody else who is launching a nutraceutical to basically and understand how and they show molecule work policies McDonalized what target does it state or a set of biological processes. So in the past we’ve done such analysis for some of the issues, people, company, physical companies, papers, and I’m licensed the research stage in order for them to launch it. But we don’t have to do that anymore because you know, pharmaceutical drug discovery, drug development space, nutraceuticals are perceived to be something like snake oil. It’s you lose credibility when you are getting into news. We can help and research, we can publish papers are using our tools, but we do launch launch articles ourselves. Some of we tried that business model ourself. Yeah.
Okay. Yeah, I saw you talking about it, one of your interviews, but maybe it was a bit of an old interview.
Yeah, it’s probably 2014 2015 but the strategy has changed. I don’t think amend the companies to get them to the interest of new coal pretty much surely. How does with a big partner.
That makes sense. So Alex has been an incredibly interesting conversation. I like to usually finish things off as a few fun questions to get your opinion on some stuff. If you don’t mind, what if anything are you doing right now or what are you taking? Is there anything that you are doing to extend or enhance your life right now or are you just kind of waiting until more significant discoveries come?
Well I’m working for in Silico medicine that’s I think that’s the best one can do nowadays to kind of stay on top of the news in general. I’m not going after a lash extension of myself at this time don’t have any protocol because right now I just focus on really hard work and the accelerating drug discovery, drug development because most of the interventions that you can basically make right now, they will have comparatively marginal differences to what’s coming up and the, I might be taking some of those molecules that may help improve performance and they can get away with sleeping less and a little bit more. Really, I do not expect substantial tangible life extension differences from benefits from the regiment I’m on right now
and with the impact that AI will have. What you guys are working on, when do you think longevity research will start having a serious impact on human life span? Not just a few years. I mean when we’re going to be extending life span by 2030 40 50 years?
Well I think we are already at the stage where AI is contributing substantially to this body of research and specifically in the AI biomarkers of aging for example in our ability to track the rate of aging in time. So now we can conduct clinical studies without the need to wait while people die. We cannot just measure the clocks using AI, so that does a huge contribution of our team and several other teams made to the longevity field. But in terms of for in general, when can we start clinically rejuvenating people and giving them a say, you know, 30 40 years of flies. I think that’s still far away. So we are talking about these couple of decades and then all of a sudden my road map is wanting to start a clinical center at some point in time and all the geography and how I want to do it, but it’s still a, at least that they see the way.
For me, I think it’s going to be quite challenging also from multiple social economic perspectives as well because the economy must be doing good in order for people to focus on something substantial long and right now people are in my off day and focus on not by the problems of hand or even inventing very short term issues to shift their attention to and that if we go into the state of the economic collapse or another great recession or greater recession, then the attention is going to switch from biomedical research, all kinds of geopolitics and internal problems that the countries have. There are so many factors of flight we can not control. I think that substantial results are at least two decades away.
Two decades away. Well, Alex, this has been super exciting conversation and honestly and Silico I’m going to be keeping an eye on it because after everything you’ve talked about in this conversation, all the things you guys are working on. Yeah, I’m very excited to see what happens with you guys and it seems like it’s going to be a very financially and fulfilling company to be working and to be running.
Well. It’s hard work but definitely it’s uh, if I, if I were to st combined, what else would they do? I don’t think there is anything else I would want to,
There was an interview I watched with you where you talked about investing in longevity is the most altruistic thing you can invest in and I 100% agree it’s going to really change the world and it’s going to change really what the experience with the human experiences. So yeah, again, I really appreciate you hopping on the call and
Well thanks for doing this. I think, you know, the more people, uh, think about longevity and more credible people think about longevity better, this world is going to be, so there is nothing more impactful. Nothing unites people more than common goal of defeating a disease, specifically feeding a disease like aging.
Thank you so much Alex.