Transcript
Speaker: From entering into a having drug approved, there is a 10% chance. The same reason why Ozempic was so expensive. Do you ever see India creating that kind of a company which creates a multi-billion dollar drug? Developing a new drug can cost hundreds of millions of dollars and only one out of 50 projects actually makes it to market.
Speaker: A big reason for that is testing. Every drug candidate has to be tested in living cells and then fixed and then tested again. Dr. Aridhneesha co-founded Immunotow AI to replace that long, expensive testing process with AI models. And in this episode of the Founder Thesis podcast, she breaks down how AI is shaping the future of biotechnology. Earlier on, we used to sacrifice the animal, get the spleen out, get these B cells, fuse them with cancerous cells to make them immortal. You inject an animal with that cell and you hope and pray that the animal develops antibodies. So the whole process of discovery where the animal was making these antibodies can now be done with our AI.
Speaker: Arindisha, you are the founder of Immunoto AI. I want to give a warning to my listeners. This is going to get technical, but I personally enjoy the technical conversations a lot more. So first of all, ah tell me what, like a big picture understanding, what is the domain within which Immunoto AI is operating? And then we'll kind of do a deep dive and break it down.
Speaker: Sure. Hi, Akshay. So first of all, thank you very much for having me on your podcast. um So Immune to AI um works at the intersection of AI and ah biotech and more specifically where we're making drugs. So it's in the intersection of AI and drugs.
Speaker: What kind of drugs? um So these are most importantly what we call as biological drugs and very specifically antibody-based drugs. um So these are biological molecules and antibodies.
Speaker: ah If you're aware about antibodies, you might have heard about it. I think during the COVID phase, there was a, you know, everybody knew about antibodies. So these are part of our immune system. These are proteins. And generally what your immune system does is whenever there is a,
Speaker: foreign agent that enters your body, right? First of all, your body needs to recognize that a foreign substance has actually entered. And antibodies are the way of recognizing. So they can bind very strongly and specifically to an invader, right? And based on this interaction, it can then activate your immune system to say, you know, there's an invader that has come into the body and let's attack it. So that's the primary job of antibodies. And I think by the end of this podcast, we'll be able to appreciate how we're using antibodies as drugs.
Speaker: So essentially, from what do I understand as a layman, that say the smallpox vaccine, I think this was the first vaccine ever developed, ah
Speaker: A vaccine is a weakened form of a virus, or used to be, I'm not sure if that's the case anymore, a weakened form of a virus which exposes the body to that virus without causing serious damage to the body.
Speaker: ah It's in a way a training run. for the body. And ah this training run teaches the body about this virus. And ah in the process of defeating the virus, the body develops antibodies. um Antibodies, you explained, are protein molecules. I don't know molecule is the right term. protein molecules, correct. Protein molecules, which... ah are shaped in such a way that they attach to the virus. So that's why you need a training run because the protein molecule which attaches to the virus needs to experience the shape of the virus so that it also evolves into a shape that can attach. um
Speaker: Once this protein attaches itself to a virus, what happens after that? how What signal goes into the body or how does the body know that this has to be flushed out? or I mean, what happens after this attaching thing?
Speaker: Okay, so um you've kind of briefly touched upon the training phase, so to say. i just want to elaborate that a little bit so that we can appreciate what exactly happens inside your body, right? So very rightly put, in the case of a vaccination or any time that the body encounters a new virus or a new particle for the first time, um the body doesn't have antibodies against that already, right? So there is an accelerated evolution that kind of happens. So the antibody undergoes a lot of change.
Speaker: And in the process, it keeps trying to bind very strongly and specifically to this target virus or a target protein. So this can take sometimes you know months, sometimes years as well. In the case of HIV, we've seen people needing years to develop good antibodies against it right So it's a process where your body's kind of doing a hidden trial that, okay, if I make this mutation, is it going to bind better or not? which is why the vaccination becomes all the more important because if you encounter the active form of the disease the very first time and if your body takes so long to develop an antibody, then you know you might the the virus might overcome and kill the person. So a vaccination really helps you go through that training phase, as you rightly pointed out, where your antibodies are developed and then stored as memory in your body. So the next time when you actually encounter a live virus, you can very quickly mount um an immune reaction and eliminate. So what happens exactly is that the antibody binds to the virus proteins. There are a bunch of proteins on the virus surface as well.
Speaker: So it mines the virus and then it activates your immune system. There are immune cells called as T cells, you know, NK cells, which can be activated. And they basically come either neutralize the virus, so kind of i surround it from altogether all around so that it cannot enter into your cells or in many cases, eliminate the virus altogether.
Speaker: right And so that's how ah in the process of of any foreign particle entering your body, this can be a virus, it can be a bacteria, it can be any pathogen, that's how the antibody helps you eliminate a disease generally.
Speaker: ah One question, you said that there are mutations and basically it's a hit and try method until something works. um Where are these antibodies being created? What is like the birthing room of these antibodies?
Speaker: So there are something called as B cells in our immune system. So these are WBCs, are the general lay term that we understand. Within your white blood cells, there are different types of cells, so T cells, B cells. And so B cells are the major factory, so to say, of these antibodies. So the B cell has an antibody on its surface, and then it keeps trying to make this mutation and attach to the target. And if it gets a positive signal, then it survives, if it doesn't bind and there's a negative signal, then that B cell dies away. So in that process, it's the the exact process of survival of the fittest, right? so How evolution happens. So the ones that can bind strongly and specifically that get um saved and the ones that don't, they die die out eventually. And that's how eventually you have cells that then segregate into memory B cells, which kind of store them as memory for life. So at any point in your life, if you encounter that virus. And that's how basically smallpox, chickenpox, these kind of vaccines work very well. um
Speaker: But this, if you might ask, but you know, your general flu ah keeps coming up every season and then we end up getting sick every single time. How come we don't have memory for that? um We do have memory, but the virus itself mutates.
Speaker: very, very fast. So every season there is a new variant of the virus and that's what happened again with COVID, right? There were new variants that were coming up. So even if you were vaccinated or even if you already had COVID the first time, there were memory antibodies, memory B cells creating these antibodies, but a new variant came about and you were not able to fight that.
Speaker: So that's what happened, which is why you fall sick every season that a flu comes in. Yep. Amazing. Amazing. This is so interesting. um How does the body know that ah this is an invader?
Speaker: Can the body also attack its own cells? Yes. So very early on in your immune system formation, right, there is a process where um ah it it happens in ah in an organ called as the thymus, where basically your body exposes um your antibodies to all of its self-science.
Speaker: So it says this is self and everything that binds to self should be eliminated so that you cannot attack yourself, right? So once that initial pool gets eliminated that these are the set that attacks self and that should not be present in the body, now anything that is apart from that set, if that enters into your body, then your body recognizes it as non-self. ah You might have heard about autoimmune disorders, right? Yes.
Speaker: Where your immune system attacks yourself. And those are the cases where, you know, a faulty ah resulting antibody that can actually attack yourself. um That is one of the ways that you can get autoimmune disorders.
Speaker: What happens in an autoimmune disorder? What's an example of one? Right. So as I said, if you're if somehow in that process of, you know, eliminating the ones that were actually binding to your tarc to your own self, um if that doesn't get eliminated, now your immune system recognizes a part of your body as foreign, so to say. the antibodies go and bind to it, and then your immune system comes and attacks it. So in that process, you do end up having um you know ah autoimmune
Speaker: disorders. So a rheumatoid arthritis is an example of autoimmune disorder. Lupus, right? So all of these where your immune system starts attacking your own self um through a faulty, ah either a hyperactive immune system or a faulty antibody, that's when you start getting autoimmune disorders.
Speaker: okay okay okay Okay, interesting. So the original antivirus approach, or ah not antivirus, sorry, the vaccination approach was weakened form of viruses exposed to the body. Is that still the gold standard?
Speaker: ah Many times, yes, but now there are much more advanced ways of not even using the live virus. So basically all that you want to expose your body is to specific proteins that are present on the virus surface, right? So you could also directly just give that protein and, you know, in ah in different formats. So there are, if you remember, again, during COVID, we gave an mRNA as a virus.
Speaker: as a vaccination, right? So mRNA is um a code. So your DNA is the main genetic code. And from the DNA, you make an mRNA, and the mRNA eventually makes the protein in the body. um And so if you directly injected the mRNA, um your body would use its own machinery to make the protein, the SARS-CoV-2 protein. um And then your body would make antibodies against it. So you can have all forms. You can have ah proteins, you can have peptides with you know um other proteins. There are various ways of making vaccines now.
Speaker: who What is a peptide? Peptide is a shorter protein. So a protein is, ah so protein is okay, 20 amino acids make up all our proteins in a very long chain. They fold in a particular fashion. And peptide is a shorter fragment of that protein.
Speaker: Okay. GLP is supposed to be a peptide, right? This ozympic big OV, these drugs. Right, right, right. The semaglutide, yes, it's ah it's a peptide. So insulin is a peptide, semaglutide is a peptide, these are all peptides. Shorter fragments of amino acid sequences.
Speaker: Okay. um What got you into this field? What's your own personal background? Okay, so ah let's go back in history. So I was born in Darjeeling and I've done my schooling to my 10th in Darjeeling. I was always interested in doing something in healthcare. So um around 15 years, when I was 15 years old, I decided to move to Delhi so that I can pursue a career in medicine. So I wanted to eventually become a medicinal doctor. um
Speaker: So, but then i couldn't get through the entrance exams for the PMT, you know, and then I eventually decided to pursue a PhD in the field of biology. So, um went on to ah ah went on to study my bachelor's in biomedical science from Delhi University.
Speaker: And then post that, I immediately ah got an integrated PhD program at NCBIS-DIFR. That is when I moved to Bangalore um and then finished my PhD. um And towards the end of my PhD,
Speaker: Till then, I was quite sure that I'd probably stay in academia, but then I wanted to do something of my own. I wanted to take up some challenging ah you know problems to solve, and then I decided to quit academia and build something of my own. So um around that time, ah Entrepreneur First had come into India and this was, I think, the fourth batch of the cohort. um Applied for it, got through, and that's where I met Trisha. I can talk at length about this, but um yeah, if you have any other questions.
Speaker: I mean, how did you identify this problem statement of antibodies? Yeah. I mean, you already knew that going into EF you discovered it or there was a process of shortlisting problem statements?
Speaker: Right, right. It was, it was. So um going into EF, I did have some ideas, but those were in a separate domain. um And while I was at EF, I ah was talking to Trisha in general. I was just like chatting with her. And then what we realized is that, you know, so Trisha comes from the AI background. She has a master's in computer science, specializing in AI. She has about, she had about 10 years of experience in robotics and automation. She was she worked at Grey Orange prior to joining EF. ah So she she comes from a very hardcore AI a background. I come from a hardcore bio background. and um
Speaker: During my PhD, I'll just backtrack a little bit. During my PhD, I worked on honeybees as a model organism. Honeybees are not very common in terms of doing research. So you have your mice, you have Drosophila, which is a fly. So you do end up getting antibodies for research purposes very easily for mice, rats, and Drosophila, et cetera. But honeybees being a non-model organism, I could never get antibodies for my ah research.
Speaker: Why do you need antibodies for research? like but So because antibodies can bind so strongly and specifically against any target, right, you can actually use antibodies and people use this very commonly in research to identify whether your protein your target protein is present or absent or where is it localized in the tissue. So you can use antibodies to do ah these kind of studies. um And I wanted to see where in the brain, and so I looked at honeybees as a model organism to study time memory.
Speaker: So there were certain genes that I had identified that are important for time memory. And I wanted to open up the honeybee brain and see where exactly are these proteins you know coming up in which neurons. And I wanted to use antibodies to visualize that. And i couldn't find any antibodies.
Speaker: Then I thought, okay, let's custom make it. And then I realized that custom making it would take a year, a year and a half. And I'll come to that in a bit, why it takes so long. A year, year and a half. And then it may or may not give you an antibody. That's another thing. Imagine you have a five-year PhD. You spend a year, year and a half trying to make this antibody. And then at the end of it, you if you realize that you're not going to get the antibody you want, that's going to be an utter waste of the time, right? So then I decided, okay, let's not do antibodies I used.
Speaker: RNA instead of antibodies to visualize where the the genes were present. So this was always there in my mind. And then when I joined yeah EF, I was talking to Trisha and then we thought, you know, maybe we can build something in the healthcare space using using AI. And then we were kind of thinking about what all problems can we go after.
Speaker: Right. And I think big data had become, so this was late 2020, mid 2020 to late 2020 and ah big data and, you know, doing a lot of data analytics had become quite popular by then. So if we were to use AI, I didn't want to do something that was the very, you know, commonly done. We wanted to have a challenging problem. And then, um we thought maybe then I knew that discovering antibodies in general is very difficult, making antibodies is very difficult. So we went about speaking with industry experts and then realized that this is a much bigger problem in the um pharma space where you want therapeutic antibodies. So there's research-grade antibodies and then there is therapeutic-grade antibodies and it's a much bigger problem there. And so that's how we then you know started this company.
Speaker: ah Both me and Trisha worked together and then we had goals of basically building an AI platform that can create antibodies for any protein target for that matter. So that's how it all started.
Speaker: the So you kind of explained to me the reason why researchers need antibodies is to... i'd identify locations of specific proteins, like say within the brain, you want to know where the gene responsible for time memory is clustered. And so therefore you wanted an antibody which would attach itself to that gene. And then you would have some way of...
Speaker: ah tracking or detecting where that antibody is probably some coloring or some radiation yeah ah yeah okay yes okay so is that the only use case on why an antibody needs to be created ah So that's what, so research is one area. Diagnostics is another area where um ah things like your COVID test or a or a pregnancy test where you have this strip of line coming up. Generally, you again, antibodies are present there. Antibody would bind to, let's say, again, the COVID virus, for example, if the virus is present. And then there is a color reaction that happens. So antibodies are extremely versatile in its usage, right? So there's research, there's diagnostics, and then there is therapeutic. So because it binds so strongly and specifically, you can do a whole lot of things with antibodies.
Speaker: How does the color change happen when an antibody binds? Okay, so there's a two-step process. So there's, let's say your target, yeah there is a first antibody which supposed to bind to this target, it binds. There is a second antibody which binds to the first antibody that we call as the secondary antibody. And that generally has an enzyme attached to it.
Speaker: This enzyme has a property of converting, let's say, a particular chemical into another chemical and the first chemical can be uncolored. And after the catalysis with that enzyme, the final product that gets formed can be a colored product. That is one way. The other way is you can have a fluorescent tag on the antibody itself. So fluorescent molecules, when you pass a particular wave light a wavelength of light, it will change it to a different color, right? So you can then visualize things under the microscope. So I i generally used fluorescence-based microscopy to look at the brains. So I generally had antibodies. These antibodies would anyway be there in that, I mean, you've fed antibodies into a solution, a very simplistic way I'm asking you a question. Yeah.
Speaker: And if the antibody binds or doesn't bind, in both cases, those antibodies with the fluorescent tag are still there. Oh, right. So that is a process of doing this. So you obviously first ah ah incubate the tissue with your primary antibody. um After, ah ah let's say, an hour or sometimes you keep it overnight at four degrees, ah then you need to do a lot of washing. You need to rigorously wash it.
Speaker: So, which is where the strong... Unbound antibodies get washed away. Washed away, yes. So, everything that's unbound needs to be washed away. And then if if your protein if your protein of interest was present, then it will remain bound. If it was not present, it won't ah be there. And then, yeah, so that's how you can visual visualize it.
Speaker: So, washing is one of the most critical steps. So, Anne, you said that diagnostic research and the third use case is therapeutic. Just talk a bit about that.
Speaker: on the on the therapy, right? So um antibodies become very important for ah diseases like cancer and autoimmune disorders that we talked about, right? Because in these cases, cancer, for example, is stemming from your own cells. um So many times your body would not be able to make antibodies against cancer with that self-non-self elimination, right? That's one thing.
Speaker: The second thing is that generally, again, cancer tends to overpower the immune system. It tends to suppress the immune system so that, you know, the cancerous cells are not eliminated. um So in these cases, now, instead of relying on your own immune system to make antibodies, if you can give antibody as a drug,
Speaker: right You externally inject that antibody into your body, um cause your immune system to recognize cancer as ah an unwanted cell type, and then eliminate the cancer cells. That's where that's one of the use cases of antibodies. um The other way that you can use is you can attach. So you know that there are chemo drugs, right? Chemotherapeutic drugs. But chemotherapeutic drugs kill every fast-growing cell in your body.
Speaker: So you wouldn't want them to kill every cell. And what you can have is an antibody that... that has this chemo drug attached to it. These are called as antibody drug conjugates. And when this antibody binds to a cancerous cell, that's when this chemo drug gets delivered only to the cancerous cell. So antibody become like very, very important when it comes to targeted therapy.
Speaker: So that's another way. And there are, again, multiple ways of using antibodies, but these are some of the common ways of using antibodies for treating cancer, for example. Is this mainstream ah antibodies for treating cancer?
Speaker: Yes, quite a lot now. um So, of course, small molecules came up much earlier. So that's obviously the first line of therapy.
Speaker: what What do you mean by small molecules? The chemical drugs, so chemo drugs are small molecules. Chemotherapy, okay. Most of your, you your paracetamol, these are all chemical compounds. These are, we call it small molecule because compared to proteins, for example, these are very, very small, right? So, um chemical drugs are generally called as small molecules. And um small molecules become your um first line of treatment generally. And today, there is a drug called as Keytruda for cancer that actually generated around $30 billion dollars in the last year. And that is used very, very ah prevalently for a lot of cancers. And the the bigger challenge with antibodies, though, is the cost is very, very high. So it's more today, you know in the first world countries, you'd probably use this more frequently than, let's say, countries like ours. But of course, even in our country, there are ah these therapies that are available.
Speaker: Why is the cost so high? A lot of factors. So first of all, it's the monopoly on that particular drug, right? When a pharma company comes up with a new drug, um they have a patent for that and no one else can come in during that period. So the same reason why, you know, Ozempic was so expensive and then suddenly as soon as the patent expired and then India can now bring down the cost significantly. That's one of the reasons. so the and And generally, pharma companies, you know, the success rate of these drugs are not too great.
Speaker: We say when a drug enters into human drugs, right, there's phase one, phase two, phase three. So from entering into a human trial to actually having a drug approved, that is a 10% chance.
Speaker: The failure rate is even high when you start right from the beginning of, can I make a drug, right? So it's, I think, somewhere between one in 50 projects actually reach the market.
Speaker: Each of these drug development can cost you millions to sometimes billions. So for a pharma company, there's generally you know out of 10, let's say one of them made to the market, they want to recover all the R&D costs, all the ah cost that has gone into the other failed project. So that's where they want to make up most of the money. So that's one of the reasons for the cost.
Speaker: um Antibodies in general, because they are biological molecules, you cannot chemically synthesize them the way that you do for chemical drugs. So you have to culture them. in mammalian cells. And then the purification of these proteins also ah is a very expensive process. So just the manufacturing, the purification, that is also expensive. So, you know, combined altogether, the costs can go as high as $200,000, $300,000. For
Speaker: um for a short Oh, okay. Okay. Okay. Interesting. um Okay. So we answered on why you would want to make an antibody. ah How do you make an antibody?
Speaker: Right. So the traditional way, which is what I said, I'll come to that later. So the traditional way of doing this is because it's a biological molecule, you inject, you do something very similar to vaccination.
Speaker: you inject, let's say, a cancer target protein into an animal. Many times, mice and rabbits are used. um The animal takes its time. It can take three months to one year or one and a half year. You keep so the same process. You give boosters, et cetera. And then eventually, if the animal does make the antibody, then you Earlier on, we used to sacrifice the animal, get the spleen out, get these B cells, fuse them with cancerous cells to make them immortal, and then that would be your source of production of this antibody.
Speaker: Now technology has advanced. You can draw the blood and from the blood, you can take the DNA of all the antibodies and you can make libraries um in the lab of different antibody DNA. You can then screen them in the lab. So these would be millions to begin with. Generally, 10 to the past, seven to nine is what the what an animal, the number of antibodies that an animal can have. And then you have to screen through all of that, identify the antibody that's actually binding to your target. These can now be as few as 10 to 20. And then you can take that DNA, put it into what we call as a CHO. There you, now that becomes a manufacturer of this antibody.
Speaker: So it you keep growing them. ah It makes these proteins and then you purify the proteins from these cultures. That has been the traditional process.
Speaker: boom Okay, okay. You said ah DNA of the antibody, would it be DNA or mRNA? So DNA is double-stranded, right? And DNA is much more stable. That's the genetic code. The DNA then forms the mRNA and the mRNA forms the protein. So generally, whenever you are making these, what we call as stable cell lines, you ensure that the DNA is there in the genetic material of the cell.
Speaker: so that every time that it replicates, because mRNA is temporary, it's made for a short period of time and then it degrades very quickly, whereas DNA is in your genetic material, right? So whenever these cells replicate, it gets passed on to the future generations and that's how you can continuously produce these antibodies at scale.
Speaker: Okay, got So ah Was your inspiration to use ah AI for ah antibody production the protein folding experiment of AlphaGo?
Speaker: No. So we started before AlphaFold came into the picture. And most people may not understand AlphaFold protein folding. If you could do a little bit of a primer on what that was and why it was so groundbreaking.
Speaker: Right, right. So um as I said, that proteins are long chains of amino acids and and there are only 20 amino acids, natural 20 amino acids. And it's always a everything that's made up in our body. Every single protein is made up of these 20 amino acids, right? Different combinations of it. in a long chain and then they fold into a 3D sort of a structure. And this 3D structure with this now you can make an enzyme, you can make, you know, you can get all kinds of functions. ah So how these long chains, long sequences of amino acids fold and what kind of folds form ah
Speaker: was something that was, you know, it's very difficult. And how we would get to know these structures ah would be through experiments called as X-ray crystallography. So basically, you crystallize the protein and you pass X-rays through this. And then based on the diffraction patterns, you can know where each of these atoms are present. And then you can build out what that 3D structure looks like.
Speaker: um and now but You need the 3D structure so that you can artificially create the protein. ah No, you don't need that. But in order to understand how this protein functions, right, until and unless you understand the shape of that protein, you know which amino acids are present where, what is it doing? Because, see, in a sequence, you can have one amino acid at ah at the first position. You can have another at, let's say, 200th position, right? But when it folds into this 3D structure, these two can come together in close proximity and do certain things.
Speaker: right So until and unless you know these 3D structures, just knowing the sequence does not really help you understand the function of that protein, how and like what is it doing, how is it achieving you know enzymatic reactions, for example. How is it converting one compound to another compound? compound So all of those things becomes very difficult to understand. so Generally, whenever we are studying proteins, if you want to get a structure, you would do things like X-ray crystallography or a cryo-EM. And that's how you would know the structure.
Speaker: um And these are very difficult experiments to do because, first of all, it's very difficult to crystallize the protein. if the protein So proteins are very dynamic in nature. They keep moving, right? And if there if there are more flexible parts, all the more difficult to crystallize it. If you can't crystallize it, you can't do X-ray.
Speaker: right ah So for the longest time, for a lot of the proteins, we don't even know what that structure looks like in reality. um And so when AlphaFold came in, red um this was first there was AlphaFold I and then AlphaFold II did much better. um We were able to use computational methods to see give it a sequence and it will tell you potentially what that protein looks like. And it was groundbreaking in the sense that there are so many genes that we were we didn't have any idea about how it looks. And AlphaFold's accuracy was as good in many cases as the experimental methods.
Speaker: right So it became ah the structures of proteins became accessible for a lot of these proteins. um So that was the whole importance of AlphaFold.
Speaker: And post-AlphaFold, a lot of companies have come about, a lot of research is being done, built on top of the AlphaFold architecture. um Now, coming to us, we started before AlphaFold came into the picture. In fact, when we started, right when we go to people, all of them were like, okay, this is a longstanding problem, never solved problem. You can't do proteins, structures using AI and computation. So there was a lot of um you know pushback that we got from industry exports.
Speaker: But I think somewhere both me and Trisha believed that this is possible. We had ideas. And and the best part of me and Trisha coming together at the ideation stage is that, you know, she could tell things that, okay, f fra with an AI's perspective, this is what is possible. And I could come with a biology perspective and say, okay, biologically, this makes sense. This does not make sense. So the entire ideation happened together. And we believe that is probably one of our rights to win in this space. right Because generally what you'll see is very either a very AI dominant ah you know company or a group or a very bio-dominant group. So having both equally, and today that's how we've structured the entire team as well. We have very hardcore scientific domain experts, PhDs and postdocs. And on the other side, we have AI, ML, engineers,
Speaker: with very strong foundations on things like generative AI. So, yeah, so that's how we started. um And as AlphaFold came about, people started being more accepting of the fact that AI can actually do you know things in protein that for a very long time was believed is not possible.
Speaker: right and Because we started then and we were building these generative models, when again, when Gen AI wasn't ah of a buzzword, so we were doing this stuff. and um Over the years, we have been able to now build foundational AI models. Everything has been built in-house, everything has been built from scratch. We do everything um ah internally and with with a very strong focus on solving this problem for antibodies.
Speaker: Okay. Can you like give me a first a zoomed out understanding? You told me how antibodies are made through the legacy approach. You inject an animal with that cell and you hope and pray that the animal develops antibodies. What is the new approach which you have made possible thanks to AI? Like a very zoomed out understanding. What exactly is it?
Speaker: Right, so if you remember, I said inside our body, what happens is there is an evolution sort of a process, a hit-in-trial sort of process to say which mutation, which combination of amino acids work and doesn't work.
Speaker: um What we have taken is a radically different approach. What we are saying is um a protein folds into this 3D structure. Now, with this 3D structure, various amino acids are coming in close proximity. Each of these amino acids now form a particular shape.
Speaker: and they have certain chemical properties, right? They could be positively charged, they could be negatively charged, um hydrophobic, meaning repelling water. So different types of amino acids come in a 3D format. So first of all, what we took was a structure first approach to say that if a protein looks a particular way um in a particular shape, that is what we need to target. So the goal is that to build antibodies that can one lock into these shapes.
Speaker: in in ah in a sort of a lock and key fashion, right? that So, now that I have a lock, can I make the key for it? One question. So, an antibody locks into a virus?
Speaker: Protein. Correct? Into a protein. Okay. So, there are proteins on the virus surface and it goes in time to the proteins on the virus surface. Okay. Now I understand. Okay.
Speaker: Hmm. so So what we wanted to build was, antibod can we design antibodies knowing that my target protein looks a certain way, has these chemical properties? um Can I have antibodies designed to go and lock into that shape and have those complementary shape and complementary chemical properties so that it can go and attach?
Speaker: strongly and specifically to that patch, right? So we are now no longer talking about making mutations and hoping that one of them goes and attaches. I am saying I know what my lock looks like. I want to just build the key for it.
Speaker: But this could have been done even without AI because you said that there is ah that crystallization method which you spoke about. No, but that only helps you if you have a protein.
Speaker: You can use X-ray to know what that shape looks like. But if you don't have an antibody, how do you get an antibody to begin with? So traditionally, you were injecting the target protein into an animal and the animal was using its immune system to make a new antibody altogether.
Speaker: Okay. and And what is the new method now or the the method which you have? Right. So now we have a target. We don't know an antibody. We don't have any idea about what that sequence of antibodies should be, what that fold should be. So now we say, if I have this target, um what should be the antibody sequence? What should be the antibody structure that can actually go and bind to my target strongly and specifically?
Speaker: So, and you cannot use alpha fold for this. This is a very common question that we get. Can't you use alpha fold, right? Because for alpha fold, you first need a sequence to begin with.
Speaker: If you have a sequence, you can give that sequence to alpha fold and it'll tell you what the folded structure of that sequence looks like, of that protein looks like. But I don't have that sequence to begin with.
Speaker: I only know this is my target and I need an antibody that goes and binds to this target. So what we do is at the fundamental level, at the atomic level, we try to understand what makes two proteins tick.
Speaker: What is it about the shapes that fit together? What is it about the chemical properties that fit together? And the reason that we need AI to do this is because um at the atomic level,
Speaker: it is a hugely complex problem, right? Because every like there are a whole bunch of atoms there. They're creating certain shapes. They're creating certain chemical micro environments.
Speaker: And it is not possible for a human to sit and say, okay, let me just have these amino acids on the other side and it will go and bind. It's not that straightforward. So we've trained our AI models to understand at the atomic level, what makes two proteins stick, both shape and chemical wise.
Speaker: Now it has learned this at the very fundamental level that it is now a universal model. So now you give it any protein. Right. It could be a virus protein. It could be a bacterial protein, human protein, really doesn't matter. You give it any protein and you say, I want an antibody that binds in this huge protein. I want it to bind only in this specific area.
Speaker: right then the AI can now create structures and sequence simultaneously. It says this should be the shape and this should be the chemical properties. And it refines each other in each other's context. And eventually as an output, we get the amino acid sequence of the antibody.
Speaker: Doesn't the antibody bind to the entire protein molecule? You're saying it binds to just one place in the protein molecule. ah very A specific patch in the protein. Okay. Okay. Got it. there's There's like a massive size difference.
Speaker: Yes. Yes. So there, ah sometimes these can be multiple proteins, right? So they're really large and there is, and from a disease perspective, you want it to bind to very specific patches. For example, if you want to, so let's say there are two proteins, there is ah a protein on the cancer and there is a protein on an immune cell.
Speaker: Okay. And these two interact with each other. And this is how the cancer has been suppressing the immune system, ah the immune cell to say, don't attack me. Right. Now, what I want to do is I want to block this interaction. I want that the cancer no longer can interact with the immune cell. So then I have to make an antibody that actually goes and locks into this patch. Right. Although the protein is probably very huge, but an antibody binding here or here is not going to do the function. So I wanted to go and bind exactly here so that that interaction cannot happen anymore. So that specificity is very, very critical. And that was another challenge with the animal system because in the animal, you have to inject many times the entire protein and the animal could make antibodies binding to you know irrelevant parts of the of the protein and you wouldn't be able to get functional drugs at the end of it. So that was another challenge with the animal system.
Speaker: Okay. Yeah, that truly was a spray and pray approach. Yes, literally. Yeah. The the example that I like to give is, you know, with that lock and key example, if you have a lock and you have a thousand keys, right, and you have and you don't know if your antibody is in that if your key is in that thousand. So you're basically just trying ah trying each and every one of them and seeing hopefully that it binds.
Speaker: Okay. Now, the way an antibody is created from an animal is you said that the blood is drawn out and ah from that the DNA is extracted and then you use the chow to...
Speaker: ah manufacture that antibody. So how does it change when your AI platform says this is the sequence of amino cells for this antibody? How how does that shorten the process? Because I'm just wondering how that AI saying sequence of amino acids gets converted into a DNA, which is what you essentially need So um from the traditional process, there were two major challenges from a time and certainty perspective, success rate perspective. right One is that the animal itself took sometimes longer and sometimes could never make antibodies for that target. right So that was one challenge. The second challenge is that um these antibodies were supposed to be in the animal system, right? So the animal's body is at constant temperature, constant pH, etc. And the protein behaves a certain way inside a living system.
Speaker: Once you brought it out of this living system and you want to make a drug out of it, it has to be stable because you're going to store them at, you know, subzero temperatures. You're going to transport it. um You're going to formulate it in very high concentration doses. So you're going to pack a lot of these antibodies in ah in a very small volume.
Speaker: Right. So if the proteins are sticky, then they are going to clump together and you know cause aggregation and all of this. So after you actually got an antibody, it was still not a drug. there was a whole process of drug development that needed to follow where you characterize this antibody, looked for stability issues, looked for aggregation. And if you found that it has these problems, what we call as liabilities, then you would again go to a hit and trial method of making mutations. Can I make a mutation and make it into a better drug molecule? So that was the second part of the challenge with the traditional method.
Speaker: Now coming to AI, right? And of course you started with maybe a 10 to the past seven or more number of antibodies from the animals creed and all of this process. With our approach now, what you can do is you can first of all make a much smaller number of antibodies with the AI.
Speaker: The AI creates these amino acid sequences. You can go back to what should be the DNA sequence that will code for this antibody sequence and you can synthesize DNA.
Speaker: and And this method of ah going back to figure out the DNA sequence for a given amino acid, this is like a legacy method. It's well known. but It's very straightforward. It's very, very straightforward. there is there ah So, you know that DNA is made up of a T, G and C, four molecules, right? And it's this combination. So, three our DNA combinations code for a particular amino acid.
Speaker: So you can very easily go back to what is the DNA code that will meet this antibody, right? That's very straightforward. um So you can go back to this DNA sequence and DNA can be very easily synthesized in today's day and age. It's it's almost like a 3D printing. So you have these four molecules and you just keep attaching them in the right sequence, right?
Speaker: So now from a digital version, you can actually get a physical version of the DNA. Right. So these DNA get printed and then it gets shipped to us, for example. And we can put this into either a bacteria or a human like an HEK or a CHO. You can put it in any of these cells and then the cells will read the DNA, make the mRNA, make the protein. and then you can purify it. So that's how we can get them ah in the real world. right That's step one. So the whole process of discovery, which you know where the animal was making these antibodies, which took maybe up to a year or more, ah can now be done with our AI. The generation takes
Speaker: hours to days, right? And so you can have the antibody sequences at hand in a very short span of time. The synthesis of DNA, if you're doing few sequences, it can be done within a week. If you're doing a large library of, let's say, 100,000 or more, then that can take about six to eight weeks.
Speaker: So yeah, in that duration, you already have the the physical antibodies at hand. Okay. ah So your ah platform ah or your model ah is throwing out the amino acid sequence and it's relatively simple to ah design the DNA from that specific amino acid sequence and then ah manufacturing the DNA or ah synthesizing that DNA is also, these are all like infrastructure which already exists. You don't need to reinvent the wheel for any of these. ah
Speaker: That liabilities problem, ah does your model solve that also or that continues as it used to happen earlier that you you will test that DNA and then figure out if it has a like liabilities problem?
Speaker: No. So we don't we don't have that problem because now I already know that I'm making a drug. I'm no longer reliant on an animal system to make that antibody, right? And I already know that certain amino acid patches can be sticky, can cause problems. So in my process of generation itself, first, it takes care that it doesn't make all these faulty animals.
Speaker: So, Inhane, your model has these guardrails to take care of liabilities. To take care of. That's one. And then even after generation, we do another stage of filtering to just ensure that, you know, anything that we feel our major liabilities, we remove them altogether. Anything that we feel could potentially be problematic, we flag them. So that we know that tomorrow, if when we go to the lab and when we do all these characterizations, if something happens, you know, ah causes a problem, we know exactly what was the cause of the problem and we can go and change that.
Speaker: Contrary to what was happening traditionally, you have no idea what the animal has given you and you're basically trying to work with that, right? And you don't even know where the problems are. So here we have a lot of control. There's rational design. There's a lot of power in your hands to say, am making a drug. I want my drug to do this function, to look like this, to have all these properties. And you can pretty much take care of all of that at the design stage.
Speaker: Okay, interesting. um I feel like maybe my next set of questions should be addressed to your co-founder. See, so far, we're talking of the model doing this and that model itself remains a black box. ah How is that model created? Is that something you would be able to speak on?
Speaker: Yeah, I can to some extent. Hopefully I'll do justice. So first of all, as I said, um we go down to the atomic level, right? We try to understand a particular structure. There is a very important, um I think one of our major IPs would be in how we, so what we get is where are these atoms in 3D space, right? Where is a carbon? Where is an oxygen? Where is a nitrogen? We get all of that information. But um There is no, you know, like what are its chemical properties? what are like
Speaker: Because of these atoms coming in close proximity, what is the effect? All of the those information is not not there, right? So we have a very important mathematical transformation which helps us convert it into a mathematical format. um one of the representations that we use is your graph-based representation, right? So graphs have nodes and edges. So every single atom becomes a node and um it gets connected with its surrounding atoms through these edges and there's message passing amongst these atoms. So what is the influence of the surrounding atoms on this particular atom? All of that gets captured and we capture a lot of the chemical, the biological properties in the node features.
Speaker: Okay, you'll have to simplify this. um I'll tell you what I understand and then you build from that. ah My understanding of a model is that, ah ah and I'll go with large language model because that's what I understand best. ah You um feed and trillions of words or sequences of words, basically all the text on the internet, you feed that to a model and the model that emergent behavior happens where the model is now able to predict given
Speaker: I am whatever and then the next word will be actually what should be the next spoken that comes in the situation. So that's how the model gets that ability to predict smartly because it has ingested so much that it is statistically able to say that there is a 90% chance that After this word, this is the next word which should come, ah you know, like nice to, then there's a high chance that it'll be nice to meet you, whatever.
Speaker: So that's how a model is made. You feed it the entire internet and it is able to statistically then decide and predict what should be the next word. Now tell me about the model which you have created. Build on this okay So you've gone to the the training and the learning part. I haven't come there yet. So one of the challenges that we have, which is not there in things like large language models, is the amount of data.
Speaker: right um As I told you, getting these 3D structures of proteins from experimental methods is extremely difficult. right so there is um I think there's only a total of about 200,000 structures. So there is a consortium effort called as the Protein Data Bank, where every ah protein structure that was ah ever you know characterized has been deposited under somebody's holding things proprietary, but most of them gets deposited in the Protein Data Bank, right? So this is accessible to everybody.
Speaker: But there's only about 200,000 structures in total. And when you look at the antigen-antibody complexes, it's, I think, today only at about 9,000 or 10,000 odd complexes. Why you using the term antigen with antibody? What does antigen mean?
Speaker: Antigen, so, okay, so anything that an antibody binds to is called as an antigen. So, this can be any protein, but, you know, it's generally anything that an antibody binds to. So, there are only 10,000 of these. Only about 9,000, 10,000 of these complexes, right? So, that's from a generative AI's perspective to few data.
Speaker: So, we couldn't use things like large language models or any of, and first of all, large language is models you can't use for our problem statement because, you know, the complexity is is huge. So um we have used, ah for example, one of the generative methods that we use is called diffusion, where you add noise and kind of make the whole sample noisy, and then you and train the model to go back from this noisy data to a denoise data.
Speaker: So this is usually used, for example, in image generation, right? So that's how the training happens. but right Now, in order to do this diffusion, and because we have such little data, we had to make sure that this small amount of data is ah very, very rich in information.
Speaker: So one of the things that I was talking about earlier, which is the graph, is one of the ways that we use to make this data richer. So every single atom is represented with a whole bunch of um features. There is a lot of ah maths, biology, physics, chemistry that goes into making this new data set.
Speaker: And this now becomes proprietary to us. Okay. So this 10,000 got multiplied by a factor of some right?
Speaker: Here, because we didn't have so much data, we kind of added all the information that we could, right? And this all came from our internal R&D. How do you represent this in a much more richer fashion so that the AI can start seeing patterns relatively sooner?
Speaker: This ah ah enrichment of these 10,000 antigen antibodies data must have also been a software-driven approach.
Speaker: um both So a lot of the scientific team sitting together and saying, you know, for every of these atoms, for example, there are 20 amino acids. Within 20 amino acids, there are these atoms and each of these atoms have certain special properties.
Speaker: right So those kinds of information was more manual and then yes and then with our um mathematical transformation, that kind of, the for example, the edge creation that I talked about, right that this atom has certain properties and it is connected with these atoms in its microenvironment. And now what is the influence of each of those atoms? That is something that would be more computational.
Speaker: Okay. right So now you give all of this rich data, every single atom to your training, to your diffusion model. right So now you first add noise to this data and then train the model to denoise this data to go back to a particular structure, go back to a particular sequence of amino acid.
Speaker: And in this process, because we do this at the atomic level, we were able to generalize this problem in the training process, right? So you keep training it and say, now if I give you a very different set of amino acids in this mathematical representation, will you be able to create a complementary amino acid sequence that can go and fit into that shape and bind to that amino acid?
Speaker: So that is how, so this is very different from how your large language models would generally work. Okay. ah Do you have a wet lab? Yes, we do.
Speaker: Why? So everything that we make digitally has to be obviously validated in the lab. And biology is one very, very tricky field. right you it's not Many times things might look very obvious, but when you actually go and start testing the things out, right as I said, the proteins are very flexible. They move around a lot. right so Although you might think that it's going to bind, it may not bind. Right. So those kind of things are there. So we do have a wet lab where we actually take them to the lab, do these validations, see if it is binding, not binding, then if it is doing the function in the cells or not. So we do the entire exercise. And then based on these information, we then kind of go back to the AI and say, maybe we should, you know, tweak things here and there. Yeah.
Speaker: Are you model company or are you a drug discovery lab? Okay. So we are a platform plus asset company, right? And what I mean by that is that we do have certain targets that we are working on in the cancer and in the autoimmune space where we are designing drugs and we hope to have them as our own asset portfolio.
Speaker: But because we also have this AI platform, um tomorrow when we are working with a pharma company and if they have a particular target of interest, they can let us know and we'll run the entire process of AI generation, experimental validation and come to the drug candidate.
Speaker: However, we are never going to open up the AI platform for a third party access. So we'll, because again, there are a lot of nuances on how these things work and just giving, ah you know, open access is not going to get us the outcome that we want. So we know how these things have to be run. There is a lot of um knowledge in how to, you know, operate things. So the AI process we'll do, we'll do the experimental validation. We'll give you lab validated real world results in what that drug should be.
Speaker: So there are two revenue opportunities for you. One is ah you mentioned that $30 billion dollar drug which is used for cancer. ah So creating a similar blockbuster, multi-billion dollar drug is one way in which you can monetize. And the second way is for...
Speaker: Other pharma companies which want to also create these multi-billion dollar drugs, ah you will provide them ah antibody design as a service.
Speaker: Correct. it's I wouldn't call it as a service because once again, um it's more about licensing. So the final drug candidate that we create, right, drug is going to be licensed out. The IP of that drug is going to be transferred to the pharma company. um Generally in this industry, A preclinical drug generally has ah triple digit million dollar deal size, and that comes with an upfront. So generally, there's a 10 to $50 million dollars upfront that comes right at the beginning.
Speaker: um And then as the drug progresses through, let's say, human trial phase one, phase two, so there are these different milestones at which the, ah you know, different tranches of cash come in. And then once the drug is in the market, that's when you would realize the triple digit million dollar value. And then there would be some additional royalties on the sales. So a preclinical drug is generally, as I said, in the maybe 200, 300 million dollar value. ah As it progresses to a phase one or a phase two, then that value goes to late triple digit to a billion or more dollars.
Speaker: Are you providing the full drug or are you just providing the antibody? the ah the The antibody is the drug, but in the final drug format. So we'll test it in um in cells, we'll test it in animals, and it should be doing the job that it should be. And that is what is going to be licensed out.
Speaker: so Like you gave that example where the chemo molecules are attached to the antibodies so that the delivery is targeted. So you would provide that full combination of chemo body plus... Okay. Correct.
Speaker: Okay. Okay. Got it. ah Why wouldn't you go all in... to this, it is, it just sounds extremely lucrative and lower risk because a you are getting something upfront every time you ah solve a problem for a client. So essentially a client is going to come to you with a problem statement that I want so-and-so to disease for which I need so-and-so kind of an antibody ah or delivery mechanism or whatever. so And then you will work on it and ah solve it. And you are pretty much guaranteed that your effort will not go in vain.
Speaker: ah You'll get at least that preclinical revenue, which you said can be as high as $50 million. dollars um So why won't you just go all in on this?
Speaker: No, we we we are going all in on that as well. But um so pharma, you know it's not very easy to go to a pharma and the pharma is not going to just, you know not give you a 10 or a 15 million right away, right? You need to show enough proof. um So, which is why currently we have about um four targets that are already in validation. They're already in the in the lab setting. We're testing out their function.
Speaker: These are your in-house targets, not for clients. These are they are the in-house targets. and um and then But then you still have to show that your AI works, right? And the bigger problem today ah for a company like ours is ever since, you know, LLMs and generative AI became very popular and then AlphaFold came in, there are a whole...
Speaker: you know, category of companies that have just built on top of AlphaFold, right? And they're trying to do things that using the AlphaFold architecture, trying to create ah new antibodies and things like that. So there's a lot of buzz and there's a lot of noise in this area. And so now it becomes even more important that you You have to show that you're not just getting binding antibodies, you have to show that you're getting functional antibodies that can function in a cell system and then in an animal system, right? So that's what we are working towards and hopefully we'll have some animal data by this year and that should really help us close some of these pharma deals.
Speaker: So whether they want to out-license the ones that we already have or whether they want to ah they want to give their own target, all of that is something that we're currently actively pursuing. Got it. ah The difference between a binding antibody and a functional antibody is essentially lack of liabilities.
Speaker: ah No. If you remember, I said ah if an antibody binds, it could potentially bind in any part of your protein. If you want it to block, for example, it has to bind to a very specific area with a certain affinity. So just binding is not sufficient. You have to actually see whether it's doing the right function in cell systems. And then, of course, liabilities is another aspect of it. But all of this put together in a package, it has to be a final process. clinical grade drug molecule and not just a binder.
Speaker: Got it. Okay. Super interesting. ah So these four ah antibody therapeutics that you are working on, ah your hope is to eventually find someone who will license them.
Speaker: Right. So, okay. So, if we had the funds, right, you know how much it takes. It takes, um as I said... Clinical trials is, as you said, like... Very expensive, right? So, if we have the funds, there are a few targets that we'd love to take it all the way through, right? But let's say if we don't have the funds, then we can either have a co-development arrangement or we can out-license the ah drug altogether. So, I think it is largely about whether we have the funds to take it all the way through.
Speaker: um The second important factor that I should mention is that all of these four are either a best in class or a first in class drug. So first in classes where there isn't any drug existing in trials or in the market, best in classes, there may be something, but what we are developing is better than what already exists, right? And um so there is a lot of potential to partner it, out license it, or if we have the funds, take it all the way through.
Speaker: Do you see and yourself getting those funds? you know there is the ah i mean, US is where pharma investments happen typically, right? That's where the the the big money is, the big investors are.
Speaker: ah Is the Indian market deep enough to give you a couple of hundred million dollars, which is what you would need ah to do the clinical trials? So ah I think the the the ecosystem is maturing, but I don't know if it has matured fully yet. ah One of the biggest challenges that we have faced is funding, right? As a company who's doing deep tech, biotech, um and no revenue for a long period of time, there's a gestation period, right? So we've struggled with getting funds and only recently we were able to close a $6 million dollar fund, whereas our competitors in the US are raising like $11 million dollars seed and then another $30 million dollars post that. Chai Discovery is one of our compet competitors they recently raised. I think they raised a $130, followed by a six month later, another $130 million. So, you know, compared to that kind of capital, of course, we don't have that capital. And so um it's good and bad, good in the sense that we have been very...
Speaker: you know, ah strategic in how we use these funds. And the whole process of having a universal platform came about because we wanted to, you know, we can't afford to generate data and then train models and then hope for an outcome. So we had to build a universal platform. So Yeah, those are the challenges. So good, but bad, of course, that the kind of validation that we are doing today after our six million raise, that's something that we should probably have done a year or two ago, right? So that scale of testing is not something that we could do in the past because of lack of funds. um So, yeah, it it is challenging, and which is why I always say if we have the funds. um
Speaker: Also, the efforts are to see if we can get US investors interested in investing. So that's obviously there. Or if we can get a pharma partner so that, you know, with some of the outlicensing, we might have enough capital to push some of our assets forward. So that's, yeah, that's we'd love to see.
Speaker: I guess there are two constraints probably. One is, of course, ah that ah that number of $130 million dollars Series A in India is extremely rare.
Speaker: um and The other is that You need a VC fund which can underwrite and drug discovery risk, which means that a fund which actually has um that kind of a scientist on its investment committee or something who can actually underwrite the the underlying technology, which again, I don't think India has those kinds of specialized funds, but probably the US has.
Speaker: Yes, absolutely. Absolutely correct. So what we've noticed is that and and with time, as I said, the the ecosystem is maturing because now when we talk to VCs, we see that there is a lot more interest in investing in deep tech and, you know, IP driven companies. But the the challenge still is that they don't have domain experts to actually kind of underrite review us and exactly underwrite. And so always, um sometimes it has happened that we've talked to a fund for six months or more and then eventually they say, okay, we couldn't bear the conviction. Right. So...
Speaker: that is one of the biggest challenges. And i have heard the stories from the VCs end as well, that they would love to have these experts. But unfortunately, you know the the ecosystem as a whole has not matured, that there are enough people who can join VCs, you know enough people who've done this in the past. So if you look at if you look at the US ah funds, right i think is one of the biggest fund there and they are like PhDs come out of Amgen and all these, you know, pharma companies who join them in in the team, right? And so they can evaluate companies much, much better. And that ecosystem just hasn't matured here yet. So it is a struggle right now. Yes.
Speaker: This $6 million dollars is essentially bet on your conviction. Like they ah may not have been able to underwrite ah the technology, but that ah the gut of a VC saying, okay, these two women are extremely smart, ambitious, and ah gu The gut of Ashish Kacholia, I should say, one person believing in a lot of deep tech companies in India. So i think um I think firstly, of course, the market and the opportunity is huge, right? That if this becomes successful, then it's definitely a multimillion to even a billion dollars sort of an opportunity. So I think that opportunity is something that's definitely there. And of course, believing in us and our technology and the fact that, again, right, that we're building foundational models. These aren't like we haven't picked something that already exists, you know, fine tuned on that. And so there is no defensibility. There's no differentiator. I don't. So I think from all of these perspectives and um I should have a special mention for Pooja, who's part of Ashish's team. And I think she has a very good understanding of, you know, she's she's a doctor herself. and she can evaluate ah biotech companies much better.
Speaker: The only success story in India in biotech is Biocon, right? Like at scale or are there others? There are a lot of pharma companies, right? um But the entire Indian pharma ecosystem is largely manufacturers and generic.
Speaker: Generics, exactly, right. there i wouldn't I don't know if I can call them biotechs as such because it's, as I said, these are chemical compounds and there's chemical synthesis in the process. And I think we've done a fantastic job in just kind of you know optimizing that manufacturing process and becoming the largest manufacturer. um But I think now the focus needs to shift towards more innovation and more discovery. And then there are efforts in the in the chemical space. And yes, if you talk about biologics, um
Speaker: Biocons, InGene, they are some of the big players in the biologic space, right? But I think we still don't have enough novel biologics company in India. I think, again, a special mention to Zumatol, I think they have an antibody drug in phase one clinical trial. It's a novel drug, first in class. So yeah, very few companies who are actually working on novel biologics.
Speaker: drugs, innovation in India, there's, I think, a handful. Okay, I want to kind of take advantage of the time I have with you to understand a little more at a macro level. ah When you say very few biologics or companies which are in that space of biologics, can you define what does biologics mean?
Speaker: Yeah, so biological drugs, anything that has a biological origin, we call them as biologics. So antibodies would be biologics. Your cell therapy, you've heard of CAR-T, you've heard of companies like Immunoact, who's built...
Speaker: ah in like um indigenous CAR T in India, right? So CAR T's are basically T cells and they have, so basically in CAR T what happens is you take a patient's T cell out of the body, you mutate it to add an antibody into that, right? And there's an antibody along with an additional protein domain. So whenever this antibody now goes and attaches to a cancer cell, this T cell gets activated and it can then ah attack the cancer cell.
Speaker: So, but here it is very, um like technologically you need, you know, a lot of things. It has to be housed in a hospital so that you can draw the patient's samples.
Speaker: make these mutants in the lab in a clean, you know absolutely non-contagious fashion and then insert it back to the human. So there are a lot of efforts in trying to make off-the-shelf card teas, basically, where you don't have to draw it from the patient and you know you could directly use it as a drug. So these are cell therapies. these would all, cell therapy, gene therapy, um antibodies, these would all come under your biologics category.
Speaker: Okay. ah What is ah Biocon's claim to fame? What have they developed? ah So, Biocon largely, I think, started with enzymes and then then I think eventually they had an insulin. So, insulin is also biologic? Yes, because it's a peptide, right?
Speaker: and And these GLPs, Ozempic and all, these are also biologics? These are like the more mainstream ah kind of biologics. Okay. Right, right, right. Anything that has a biological origin, you can call them as a biologic. So, um Yeah, I think so. Insulin is probably their ah biggest blockbuster. And then they were also doing services. And then eventually, I think they kind of ah separated out Syngine. So Syngine is now only services. Services means that outsourced drug development.
Speaker: Yeah. So let's say let's say a pharma company can come to Syngine and say, I want to develop an antibody for a particular disease. And they would do the animal immunization, characterization, everything as a service. of Then i think also there is ah another spin-off from Biocon called Bikara, which has, again, another drug which is in trials ongoing, but I think they're registered in US if I'm not wrong.
Speaker: Why is ah insulin a big deal? I thought insulin is like a commodity, like millions of diabetics use insulin, which is relatively affordable.
Speaker: So, like, but why was that a, was it a hard drug to make that, I mean, you were praising Biocon for cracking insulin. So I was just wondering that, ah you is it, why why why is it praiseworthy?
Speaker: I think at that time, right, at at that time, today technology has advanced way ahead, but I think at that time, probably it had a huge impact. And I think it was again similar, right, whether you can get this as cheap and as economically as, you know, as other drugs. um So I think today, i think today it's not so much of a problem, but there was a time when, you know, just making these cultures biologically was probably difficult.
Speaker: And ah these companies which discover drugs, do they also sell? Because selling is a different DNA, right? You need to have that army of medical representatives and that whole sales and distribution machinery. So someone like a Biocon and these kind of companies, are they also have they also built that sales mechanism or do they typically license it out?
Speaker: No, i think I think Biocon does its own. I think they are based not only in India, also in a lot of countries outside India. um But yes, you're very right in that sense that even in the in the drug space, right? Pharma generally, if you look at the big pharma, um they are mostly the the sellers, the manufacturers and the sellers. They're like a Walmart, like they're the distribution machine.
Speaker: Correct, correct. And they have all these channels to to kind of ensure that the drug goes into the market at the right pricing. And and there is an army of MRs, as you're right you pointed out, who kind of go and sell this. um whereas ah Whereas drug discovery is a whole different format. So...
Speaker: Yeah, I mean, you will need that. And which is why a lot of the pharma, um while they do have in-house R&D, but many times, um depending on each pharma, they may have a 60-40 ratio of in licensing versus developing developing it in-house. So, you know, there's there's all of that. So depends on every single pharma.
Speaker: Now, when you are ah hoping to create your own therapeutic antibodies, which is essentially a biologic or a drug, um do you see yourself actually getting into that whole sales and distribution or you would license it out to a pharma company?
Speaker: Yeah, I think if yes, that is probably much later, not anytime soon, because again, as we discussed, right, that's a whole different ballgame. So to the extent possible right now, we'd like to stay focused on getting new drugs. And then about when do we out license it? That's something that can be dependent on the funds, whether we take it to phase one, phase two or phase three. But I think for the ah for the near future, it is going to be out licensing.
Speaker: Yeah. Only once we build those capabilities is when we can start ah distribution as well. Yeah, that's different. And I'm guessing these pharma companies themselves are like venture capitalists in a way. Like if they pay you for a drug which has not gone through clinical trials, they are in a way doing that venture funding ah because they're taking a bet that this will cross clinical trials. They have that expertise to underwrite it.
Speaker: Yes. And there are there are different formats as well. The pharma companies themselves also have venture arms. And sometimes they do research collaborations. They do take equity. And so there are different ways in which these pharma also can kind of work with you.
Speaker: When a pharma company comes in for a drug which has not gone through clinical trials, what are they signing on? They're signing on that we own the license to distribute manufacture and distribute this drug. The IP, basically, we'll have to transfer the entire IP to them.
Speaker: Okay. And then they're going to... And then the risk is on their books in terms of whether it clears clinical trial. And the cost of clearing clinical trials is also on their books. All of that on them. And then we, as ah as a benefit, what we would get are these...
Speaker: you get some some sort of IP sale money that you can walk away with, but it would be relatively like a small percentage as compared to what you would have got if it had crossed the clinical trial.
Speaker: Yes, absolutely. Okay, so ah let me kind of wrap up by asking you this. you know ah That dream of creating a billion-dollar drug or a $30 billion dollars dollar drug is only possible if you are getting a couple of hundred million dollars as funding. Do you ever see India creating that kind of a company which creates a multi-billion-dollar drug?
Speaker: Yeah, I think ah ah definitely, I think we are headed there. As I mentioned, I think the ecosystem is is at least, I mean, there's a attention that is being given to this. um um there are There are some government funds and and schemes that are coming up in the pharma, in the in the biotech space, right?
Speaker: However, one of the challenges that we face at the ground level is that many times it's very, so the the quantum of the grant, for example, is smaller, relatively smaller compared to what kind of money we actually need to do um drug discovery sort of challenges.
Speaker: How does it compare to the money the government spends on semiconductors? Like there is some sort of government incentives and grants, etc. for semiconductor development. Right, right. No, I think the semiconductor space, because again, right, because it started ah slightly earlier, and I think now you have a whole ecosystem that is being built, I think there are a lot more opportunities there. And and that is the problem with things like this, right? Where semiconductor, biotech, space tech, these are domains where there is a long period of time before you can actually see, you know, revenue and commercialization. And in these sort of domains,
Speaker: ah the government schemes and grants could be extremely helpful in kind of, you know, doing that R&D, in building things out, building novel, innovative, IP-driven products, right? So government support and and again, while while the government is doing things in this domain, right? Some of the problems that we face, as I said, is the quantum number one is too small. the time it takes, again, given a startup that's you know competing with with global companies out there, um the time that it takes for these grants to come to you is sometimes very long. um and And in general, the decision-making process, we believe, it has to be very outcome-driven, that there is science and this is translatable into something that can commercialize. Right. And so that is the kind of support, the infrastructure, the ecosystem and funding. Funding becomes the biggest challenge. I mean, ah we know about some of our competitors who got like a 2 million euro grant. Right. And we can't imagine something like that here today. So
Speaker: We need those kind of quantum of money to actually get to making these multi-billion dollar drugs. So there are efforts, but I think we need to deep dive into ensuring that it goes to the right people and enough money, enough quantum of money is going into these companies so that something big can come out. So I think we're seeing space tech now becoming much more you know mature in that sense. We are seeing semiconductors. And now I think it's time that we invest more in the in the biotech and the drug discovery sort of space.
Speaker: What would the change attitude? Is it that we are one success story away? That if there's one big outcome success story and that will change attitudes? or Or is it a realization that you need sovereign technology? Like that sovereign tech is a big theme these days, you know sovereign AI, et cetera, et cetera. Is that going to change the attitude?
Speaker: Yes, I think, of course, like success to it. OK, so one thing is, I think we as ah as a country, we are very risk averse, right? We don't like to take risks. And the amount of R&D funding, even within private companies, right, even within pharma is...
Speaker: very, very small compared to what you know countries like US are doing. um So I think that ah we have to be more risk taking in our attitude and and be able to, because see, eventually with services, you can do a certain extent, but if you want to unlock true value, right it has to be IP driven, it has to be innovation driven. And and you know you're no longer reliant on somebody else. right the whole If you remember um how we started making generics is you know because there were these patents and they were the excruciatingly high costs of these drugs. And then eventually India took a stance and said, you know we want to make these drugs. And there were these patent laws that were kind of amended to ensure that India could make ah generics, for example. So you're saying India diluted intellectual property protection so that Indian companies could manufacture drugs for Indian populations? I think there was a whole ah you know um process where not really diluting, but you know there were certain things that could to be patented and there were certain things that could not be patented in in that process of making drugs, for example. So, um and it it has gone through some ups and downs. And and so I think there was ah there was a phase when you could you could pretty much get away with patents by just doing a process change, right? So I think the example is
Speaker: A plus B makes C, and then somebody could say that, okay, we did an A plus B, and that D was pretty much B in some sense, right? So you would get away with it. And then I think there were further amendments to ensure that that protection came into place. so Yeah, so we have enough.
Speaker: Yeah, we did we did get past that fact that, you know, a big pharma could always hold a patent and we weren't able to make them. I think we did get past that stage and then we started making generics um in India. And that's how the entire boom of ah pharma generic manufacturing happened in India. So in a way, ah this success in generic manufacturing is also...
Speaker: possibly ah can be the reason why there is no investment for IP because I guess the the government, ah from what I understand, while you didn't explicitly say it, but it does seem like IP protection would be stronger in a US market as compared to India and therefore the value of IP and the ability to fund IP in India would be much lower.
Speaker: ah i I don't think I can comment on that specifically. Yeah, okay okay. Got it. So, ah yeah, but... Yeah, but in some sense, I think there was a lot of value creation by just manufacturing these drugs, right? And that is where a lot of the focus went. And then after the genetics came, the biosimilars, when antibodies and biologics started coming up, then the genetic version of biologics was biosimilars, right? So that's where, again, Biocon started.
Speaker: ah comes into the picture there are other companies like Enzyme so we're all like making these biosimilars but what about actual novel drug candidates and there's a lot of risk that is associated with it and for some reason we've always been risk of us amazing thank you so much for your time with me I really enjoyed this conversation Thank you so much. Thank you for inviting. i had a lovely ah time and I think it's been quite some time that I've, you know, spent time talking about the basics of why and what we're doing. So it was it was a fun conversation. Thank you so much
