Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Disrupting Biology With Algorithms | Dr. Manoj Gopalkrishnan @ Algorithmic Biologics
214 Plays3 years ago
This episode is a crash course in molecular biology, a $62 billion industry, and its several real-world applications.
Akshay Datt speaks with Dr. Manoj Gopalkrishnan, Founder and CEO of Algorithmic Biologics, a deep-tech startup providing the highest quality molecular testing at a fraction of the market cost. Their first product, Tapestry, is an award-winning software delivering affordable large-scale Covid-19 testing.
Dr. Manoj is an alumnus of IIT Kharagpur who, after completing his doctorate in the US, came back to India and became an associate professor at IIT Bombay, before starting Algorithmic Biologics.
Know about:-
- Contributions during Ph.D. and RSA algorithm
- RT-PCR test and Tapestry
- Challenges of the medical diagnostics market
- How molecular computing leads to better molecular testing
-----
If you want to be a podcaster, then check out Zencastr.com which is also the generous sponsor for this episode. Zencastr is like a Shopify for podcasters taking care of all your needs from recording to editing to hosting.
Show your love for this show by using this link to sign up: https://zen.ai/founderthesis
Recommended
Transcript
Introduction and Sponsor Shoutout
00:00:00
Speaker
Before we start today's episode, I want to give a quick shout out to Zencaster, which is a podcaster's best friend. Trust me when I tell you this, Zencaster is like a Shopify for podcasters. It's all you need to get up and running as a podcaster. And the best thing about Zencaster is that you get so much stuff for free. If you are planning to check out the platform, then please show your support for the founder thesis podcast by using this link, zen.ai slash founder thesis.
00:00:27
Speaker
That's
Manoj's Academic Journey
00:00:45
Speaker
Manoj is probably one of the most academically qualified founders we have featured on the show. He is an alumnus of IIT Kharagpur who went on to complete a PhD in the US and then came back to become an Associate Professor at IIT Bombay. Early on during the pandemic, he created a technology to bring down the time and money needed for testing millions of Indians for COVID.
00:00:55
Speaker
zen.ai slash founder thesis.
00:01:08
Speaker
But very soon you realize that to really make a difference with his technology, he would need to wear the hat of an entrepreneur in order to make his technology into a commercially viable product. This conversation is a peek into the mind of a scientist who gave up the lab for a boardroom and it is packed with nuggets of knowledge about molecular biology. This is the type of conversation that will make you smart enough to talk with the best minds in biotechnology. Listen on.
00:01:34
Speaker
So what happened was after my third year undergrad, I got the opportunity to do a summer internship at Georgia Tech. And so I got to spend two months there at Georgia Tech doing some research. So they gave me a library card. They gave me unlimited access to the library. Read as many books as you want.
00:01:54
Speaker
I literally, I swear, I went in there on the first day, and it was amazing. I could find whatever book I wanted on the computer, I would go where it's supposed to be, it was there. I think that's a lovely library, but this was not the case. It was not as well organized. I staggered to the desk with 30 books in hand, literally 30 books in hand.
00:02:16
Speaker
like no connection to each other, like all kinds of really, really strange, really wild things. And I was only asked one question, would you like a bag with that? They gave me a bag. I think that is when I knew that I wanted to be in a place where there were libraries like this, at least for a time so that I could exhaust this
PhD and DNA Computing
00:02:36
Speaker
thirst. So I got a full score on my GRE. So I got a full score in verbal analysis. Full score means like 100 out of 100.
00:02:43
Speaker
Yeah, this was like 2400 in my time. So it was 2400 out of 2400. So so and I had the rest of it was also good. So I could sort of have had the pick of which university I went to the one university where I got in my eventual advisor, he probably doesn't know this story because he never replied to my first email.
00:03:01
Speaker
That I was going to leave after a Masters. So he gave me an admit I went there and in my mind I was still going to leave after a year. He was an amazing guy. He just won the Turing Award one year back. So I mean it was a privilege working with him. What does the Turing Award recognize?
00:03:21
Speaker
So the Turing Award is to computer science, what the Nobel Prize is to the rest of the sciences. So it is the biggest award given in computer science. And for him, he got it for the RSA algorithm. He is the A of RSA. He is Len Edelman. What is the RSA algorithm?
00:03:41
Speaker
Oh, so any place where you're doing a secure transaction, say for banking, there is an algorithm running in the background which secures that transaction and so that algorithm, at least a component of that algorithm is going to be this one, the RSA.
00:03:59
Speaker
It is encryption. He got very interested in viruses. He started studying HIV. And then he came up with a theory about HIV, about how to combat HIV. So he said, I have to prove this theory. He took a sabbatical year, went to a molecular biology lab for a year, and then just worked on learning molecular biology.
00:04:15
Speaker
During the course of that sabbatical year, he came up with this idea called DNA computing, this idea that molecules of DNA in a pot could actually behave like a computer and you could program them to solve problems for you. So that's how he got into the lab. That's how he started in that direction. Now, when I applied to him in my head, I was more, I was a theoretician. I was interested in computation and what else could be computation.
00:04:40
Speaker
unorthodox computation. So I thought I will go there and I will do my theory thing. But after going there in the very first week, he pushed me into the lab and I was cooking with DNA and I was making nanostructures and I was imaging them with atomic force microscopes. So I was in the lab. So before I knew it, it's like I didn't know which way the world was turning and I was doing chemistry. What is unorthodox computation?
00:05:07
Speaker
So, we are used to thinking of computing as a Boolean, right? 0s and 1s and AND gates and OR gates and everything built out of that. So, this paradigm is something that we have gotten used to so much that we kind of lose, you know, we lose sight of the fact that computer science predates computers. Computer science was invented before computers.
00:05:30
Speaker
It was invented by logicians, Alonzo Church, Kurt Godell, Alan Turing, and in a purely mathematical framework. And the ideas of computer science are very abstract. And one manifestation of those ideas is the modern computer, the Boolean computer. But there are many other ways to manifest those ideas.
00:05:50
Speaker
You might imagine that water flowing in pipes could do computation. You might imagine that some kind of slime mold growing might compute shortest paths. And so all kinds of opportunities exist for viewing phenomena. What's happening in the stock market has some kind of computation. What's happening in economy at macroeconomic level has some kind of computation. What's happening in a liquid when it's solving the Navier-Stokes equation has some kind of computation.
00:06:18
Speaker
So I was interested in quantum computing, what's happening when various pins are entangling as some kind of computation. So I was interested in this. I was interested in understanding what is computing in that more abstract way and seeing where it manifests. And so I think that was what pulled me to the kind of programs I applied to for my PhD.
00:06:40
Speaker
I think I had a bit of romanticism for whatever reason, right? I had this in my head that, yeah, I want to do this. And again, right? I did not think I was committing for a long time. I did not think I was committing for life or for even five years. In my mind, it was one year do this and yeah, then go do something else. So yeah. So what was your, like at the end of your PhD, what was that one field in which you were the best in the world?
00:07:07
Speaker
So, I had worked on a couple of problems. So, I started off with DNA self-assembly. So, I built nanostructures that could make triangles and hexagonal lattices. I built nanostructures that could fold into cylinders and mobius strips.
00:07:24
Speaker
and all out of DNA. I worked on chemical reaction networks and so these are systems of differential equations, ordinary differential equations with polynomial right hand sides and it's an entire class of systems. So it's not just about one system but about the entire class of an infinite number of differential
00:07:46
Speaker
equation systems and so proving theorems about them and Making sense of them and at that time I did not know why I was doing it except that my advisor was interested in it So I was proving theorems. I got into the math of it. I got very excited by the math I sort of confronted an open problem which was which is now open for 50 years and I
00:08:07
Speaker
made some original contributions to it during my PhD. And then later after my PhD, I proved a special case which to date it remains the strongest partial result for this open problem. So that was all math and differential equations. And the third thing I did was I was thinking about physics. I was thinking about how much energy is required to do computing.
00:08:31
Speaker
Does a clock require energy to be accurate? Things like that. And again, I thought about it a lot, had lots of discussions, wrote a lot, but never published anything at that time. Again, some of these things saw the light of day several years later when I started my faculty position and some of these ideas crystallized and I could actually write about them in
Post-PhD and TIFR Experience
00:08:52
Speaker
a meaningful way.
00:08:52
Speaker
So I would say I was the world leading expert in that branch of mathematical analysis of chemical reaction networks. And I had no idea why I was doing it. It was purely the intellectual fun or the satisfaction of solving it like solving a problem. So all I knew was people cared about it.
00:09:14
Speaker
And if I solve that problem, then people would know that I was smart. So I think that was all that counted. Yeah. Okay. Okay. So then what, like once you received your PhD, what did you do after that? So once I received my PhD, I came back to India, to TIFR, where I was offered a faculty position. So I started a faculty position there. Again, romanticism. I think I saw this Institute by the sea and what's the full form?
00:09:43
Speaker
DIFR is Tata Institute of Fundamental Research. It is the institute that Dr. Homi Bhava started and it was responsible for the beginnings of India's nuclear program. It has remained a fairly small institute, fairly exclusive and DIFR and ISC are considered the, you know, top scientific institutes in India. So, it was a good position, not much teaching. They don't have
00:10:08
Speaker
in Mumbai in Colaba in sort of the southernmost tip of Mumbai. So you can see the governor's bungalow on the other side of the sea from there. Marine Drive is also visible. So it's a very nice exclusive place next to Navy Nagar. So I think the romance of it was kind of and it was Bombay, right? So I got a chance to come back and I had never intended to
00:10:29
Speaker
live in the US that had not been in my mind. I was going to the US was just to make full use of libraries and come back. So yeah, yeah, so yeah, by the way, I did manage to do that at some point during my PhD. I think I had some 236 books checked out from the library.
00:10:50
Speaker
And nobody said anything, they didn't have any problem. Okay, so and TIFR was largely about like being a faculty mentoring students and things like that. Yeah, so TIFR doesn't have an undergraduate program, just, you know, PhD program and a few masters students in disciplines that require master's students. So yeah, that was all research. It was just me talking to myself most of the time, occasionally a couple of students around.
00:11:20
Speaker
Yeah. And what field again, same field or like? So, so I, so the reason I took this job up is also interesting. So I had the option, of course, after finishing my PhD of whether to stay in academia, whether to go to industry. And I had never committed to a life of research, at least in my mind, but I hate leaving things incomplete.
00:11:42
Speaker
And so this problem was hanging on my head, and I thought I could solve it. Another problem I had been thinking about, I thought I could make progress. So I just wanted a place where nobody would bother me, where I could just go and do this stuff. And I didn't want to do anything new. I just wanted to finish this and wrap it and go and do something else. So that was my motivation. So also one reason to not stay in the US, because if I were in the US, I would have to write grants. I would have to justify why I was working on what I was working on.
00:12:12
Speaker
And that would probably move me away from this to working on something else, right? Because funding is competitive there. So here it's nobody to say anything. I just do my own stuff. Yeah. And this unfinished problem, you're talking about the physics problem, like about both of them, the physics problem and the mathematical problem. So both those problems I worked on during my time at TIFR, I made progress on both and finished part in the mathematical problem.
00:12:42
Speaker
What was it that you proved? I had proved a small special case and I thought I could prove the full thing. And so I came back to TIFR. Then in between, I went to Duke University as a research assistant professor in the math department there, spent a semester there with a couple more collaborators.
00:13:02
Speaker
And we proved a very strong special case of that problem. So that remains the most advanced result on that problem to date. What did you solve in the physics problem about the power needed for computation? So first of all, it starts with a paradox in the sense that the existing results said that you don't need energy for computing.
00:13:26
Speaker
So essentially you could do computing for free, which is counterintuitive, but they were well argued results that you don't need energy for computing. But what, I mean, what does this mean that these results, what results are you talking about?
00:13:41
Speaker
So, Charles Bennett was one of the major figures in this. He was a researcher at IBM and so he had written some papers arguing that computation can be done in a thermodynamically reversible manner without spending any energy for it. Okay, so this is purely theory, but his theory was solved.
00:14:02
Speaker
That's what you're saying. This was theory. His theory was sound. And people tried to take that theory to practice. In fact, IBM itself had spent some several hundred million dollars on trying to build computers out of Josephson junctions. So superconducting coils. This was in the 1980s. So this was a separate project from the silicon project, the PC project.
00:14:23
Speaker
And they had a team of 500 people that one day they just fired everybody. They stopped the project after they realized it was not working. They didn't realize why it was not working. They kind of had some ideas which were, but I don't think they quite put the nail into why things like this don't work or didn't work.
00:14:41
Speaker
So that was where the thing was and people were still some people were still trying to use those ideas to build real computers. So, so I think my realization was that his results were so physicists often tend to make spherical horses, right? You make some neat assumptions to try to do your calculations for you. And so his so he had been talking about the energy required to do computation if you had infinite time.
00:15:06
Speaker
Okay. And right and of course, that is something that once you wise on to that, then who wants computing after infinite time, right? We want to talk about finite time limits, then the nature of the term which dominates is different from the term that he had analyzed carefully.
00:15:25
Speaker
So then the whole thing changes and then it starts to become a question of how will you keep track of that term? And is that a term that is something fundamental or is it something that depends on technology? So is there some kind of a limit where I can make my technology better and better and better so that that term becomes smaller and smaller and smaller or is there something fundamental?
00:15:48
Speaker
And so this question I sort of approached in an axiomatic manner. I came up with a cost function. I said that this should be the cost function. And from that cost function, I showed that there is a cost. There is still a gap in my story on that in terms of justifying that axiom. The answers are pleasing. The answers that I get out of this process seem reasonable and in keeping with some observations of the world around us. But that core axiom still remains to be validated.
00:16:18
Speaker
And what is that core axiom? So that core axiom is essentially about the, so you pay for a perturbing systems. So if a system is already going to travel a certain way and you want to perturb it very slightly, you pay a little bit. If you want to perturb it more, you pay more. So somehow there is a quantitation. So you pay more, the more you have to perturb a system. If you don't need to perturb a system, you don't pay for it.
00:16:45
Speaker
So it's something like that. And it comes from a field called control theory and within control theory, cost functions like this are called Kulbagh Leibler cost functions. And so I'm applying those kinds of mathematical objects to the analysis and the answers come out nicely, the calculations work out nicely. And so my belief is that this is somehow tied to the thermodynamic cost and that this is a way of
00:17:09
Speaker
viewing the problem.
Transition to Entrepreneurship
00:17:11
Speaker
Yeah. So what then after TIFR like, and why did you decide to move on from TIFR? After TIFR was IIT Bombay and TIFR was a tenure track appointment and I didn't make tenure. So essentially they didn't give me tenure. I had published a few papers, which I was very proud of, but I couldn't make them see it the same way. So at that point, again, I had this kind of choice in my life.
00:17:37
Speaker
whether to stay in academia or whether to go to industry. I actually got a job with Amazon Seco team in Boston at that point, which was very tempting. In like data science or something like that. Yeah, in data science. Yeah. Yeah.
00:17:53
Speaker
And at the same time, I had interviewed with Google's DeepMind team. I had interviewed with IIT Bombay. And so I got IIT Bombay. Google DeepMind, I didn't get an offer, but I really enjoyed. I flew to London and I sort of saw what they were doing up close. And I got excited by what they were doing. And I realized that there was a connection to what I had been doing.
00:18:17
Speaker
at least at the mathematical level. And that, again, that bug bit me that I have to flesh this out. And so I, at that point, again, it wouldn't let me take up Boston job. I had to work out those ideas. So again, that monomania, right? I mean, so I think that monomania continued and then I just wanted to finish that.
00:18:36
Speaker
So, so I joined IIT Bombay for that reason. And here in IIT Bombay, I wrote some papers showing how chemical reaction networks could do machine learning. So instead of neural networks, you could, so taking my advisors idea, right, that you can do computing in a pot. I took it to machine learning that you can do machine learning in a pot. And so we describe networks of reactions, which could probably do various kinds of machine learning algorithms. And so that took up that phase of my life after I joined.
00:19:09
Speaker
machine learning is essentially when you feed a lot of data and then the system figures out relationships between that data to make predictions, right? Well, that's how it is taught, but that's not necessarily how it's always been thought of because for long periods when machine learning has been
00:19:32
Speaker
invented and practiced, there hasn't been a lot of data. So there are also ways of thinking about machine learning when there isn't a lot of data. So what I would say is, so for me, the most significant algorithm in machine learning is the expectation maximization algorithm, which goes back to statistics. And I like to draw analogy to Western philosophy and its development.
00:19:54
Speaker
in terms of Leibniz and rationalism and then rationalism and then empiricism, the empiricists coming from the other side of the channel. You are assuming I understand rationalism like it's empiricism, you'll have to break it down.
00:20:13
Speaker
So, the big question in knowledge, what is knowledge? How can I know something? This was the big question in Western philosophy. And so the continental philosophers, Leibniz, Descartes, and so on, they said that there are some statements which are self-evident, the axioms.
00:20:34
Speaker
And if we start from these self-evident statements and proceed in an orderly manner, then we can be assured that we will remain within what is true. And so this is how you can find truth. That was that prescription.
00:20:50
Speaker
On the other side of the channel, the empiricists, so David Hume foremost among them, countered this saying, how can any statement is self-evident? Many things seem as if they are true. You can say it's self-evident. For example, you can say earth is flat. You can say sun goes around the earth. But if you think about it carefully enough, you find that these are all falsehoods. So how can you start your discovery of knowledge standing some that is not solid? This was also about religion in many ways.
00:21:18
Speaker
This was also about whether you believe the Bible and things like that. And so David Hume was coming from more of a protestant place. So there were also a lot of those social things behind the background. So David Hume's prescription was anything that we know has to come to us from experience, has to come to us from sense perception, from observation. And so Francis Bacon and the Scientific Revolution.
00:21:44
Speaker
putting empiricism before everything sort of tied in with this again, right? It happened. Francis Bacon was on that side of the channel, same as David Hume and so on. So the resolution for this in terms of Western philosophy came with Emmanuel saying that these two are really two sides of the same coin and that you cannot, the empiricist can also not discover knowledge because if the empiricist is getting data,
00:22:09
Speaker
How does the empiricists know whether to believe that data or not? How does the empiricists know anything at all? Empiricists can also not know anything. So the only way to proceed is to posit something and then use what you have posited to make sense of your observations. Use the observations to refine your hypothesis. Use the hypothesis to now do a new test. And so it's a conversation. It's a dialogue between the rationalist and the empiricists. So this was Kant's synthesis of these two strands of thought.
00:22:37
Speaker
And so the EM algorithm is an algorithmic way of doing the same thing. It says you have an idea about the world. What is the EM algorithm? Expectation maximization algorithm. So it is an algorithm that was proposed in I believe the 1970s by two statisticians. And it is one of the most, it's there all over the place. Any algorithm you pick in machine learning, if you look at it carefully enough, it's probably a manifestation of the EM algorithm or at least a part of the EM algorithm.
00:23:07
Speaker
So the EM algorithm turns out to be, yeah, so I like to think of the EM algorithm as a manifestation of Kahn's argument in the world of mathematics, in the world of algorithms. And so what it says is you start with some data, you start with some hypothesis, you see how well your data fits the hypothesis, and you change your hypothesis to fit the data better.
00:23:30
Speaker
And once you have your new hypothesis, you again observe the world, you get fresh data and you interpret it again. And so you do this back and forth, you do this conversation, but in a mathematical manner and you kind of have a ratchet in your knowledge. So your knowledge is increasing, it's going in one direction. So that's the idea. Like the level of confidence on the hypothesis keeps going up.
00:23:57
Speaker
So, you become less wrong. I think that's the nicest way to say it. You become less wrong through time. And so, I think that is the point here that once one thinks of machine learning as a bootstrapping process, as going from a place of less knowledge to a place of more knowledge, then you can ask, can a bunch of molecules in a pod do that? Can they become more knowledgeable about the world somehow?
00:24:24
Speaker
And if you can come up with an algorithm, if you can sort of write down a bunch of reaction networks and prove that they implement this algorithm, then you're saying, yes, they can. And then you start wondering if this could happen. This could be related to origin of life. This could be related to what happened before the origin of life and so on. So I don't have answers to those, but those are also things I think are very related.
00:24:47
Speaker
If you like to hear stories of founders, then we have tons of great stories from entrepreneurs who have built billion-dollar businesses. Just search for the founder thesis podcast on any audio streaming app like Spotify, Ghana, Apple Podcasts, and subscribe to the show.
00:25:07
Speaker
Okay, okay, so this like machine learning in a pot was not essentially something which would deal with data the way we understand it like numbers or whatever but it was more of a exercise to see if Molecules have the ability to learn and no more. Well, it was so it was Yeah, so so the project that my advisor had started with DNA computing and
00:25:35
Speaker
was essentially about seeing how one could build computers out of molecules. And also there was a parallel project there.
00:25:45
Speaker
how can we understand how the cell works right. So, the living cell is a remarkable object remarkably sophisticated at the chemistry level it is just a bag of molecules, but it is not like any other bag of molecules it is it can self replicate itself, it can provide blood cell can chase a bacterium and so on right. It is a smart bag of molecules.
00:26:07
Speaker
Yeah, it's a smart bag of molecules. So where does that smartness come from? What are the algorithms running in there? This has been a question for a long time, right? So just like for the brain, we want to understand how does vision work? What are the algorithms running in the brain? For a cell, we want to work. We want to understand what are the algorithms running in a cell that are responsible for its sophistication, for its behavior. So this was what you were doing in IIT, in your tenure there, in addition to being like a faculty member and
00:26:36
Speaker
maybe mentoring the PhD students or something like that. So then what next? Like what made you finally commit to leaving academia? So yeah, I mean, I got married.
00:26:50
Speaker
And so my wife was an MBA and she couldn't relate very much to the kind of stuff I was doing and couldn't see the value of it. And I was living in an ivory tower, right? I was pursuing curiosity for curiosity's sake. And I think getting married and then having a kid later on really grounded me in some ways.
00:27:12
Speaker
and I became answerable. I had a boss for the first time in my life. Then I had to start thinking about how this is going to impact and where is this going. She was with a startup, so I was seeing her startup journey and that was quite exciting. She was with a startup called Practically, which is in the education technology space. They are creating AR-VR content for education.
00:27:40
Speaker
So I was seeing that journey. I was sort of backseat driving their startup. And so yeah, then COVID the pandemic struck. And also I had, I started following a little bit about biotechnology entrepreneurship and what's going on.
00:27:56
Speaker
I started keeping an eye out because I also felt a field has a time and I felt that this field of molecular computing had reached a certain maturity and ideas had come together. One or two companies were already out there building things like DNA memory for the cloud.
00:28:13
Speaker
So DNA is an amazing storage medium because incredible information that in a shoebox you can have as much memory stored as in an entire farm of servers. So an entire data center.
00:28:28
Speaker
So some companies have started recently who have been building workable DNA memory solutions and have started selling it commercially. And this was exciting for me because I thought if they can do this, then there are so many other ideas in molecular computing. Maybe the time is right. So I was kind of plotting and planning. I was thinking maybe this is the time now for me to finally step out of academia, maybe get a job.
00:28:52
Speaker
And then I'll find the right co-founder. I'll start something in two, three years. Of course, everything became fast forward in COVID because the pandemic struck. There was a need for more testing. I had some ideas that could help. So I jumped in. I got a lot of support from colleagues, a lot of support funding from IIT. And we built something. We were finalists in the XPRIZE competition for COVID testing.
00:29:18
Speaker
And so things go very fast there. But tell me the details. What was your thesis? What did you see as the gap in the current tests that you felt you could improve on? How did you build something better, like the beauties and the details?
00:29:35
Speaker
Yeah, so there was a shortage of tests in the early days of COVID and different countries handled it differently. In the US, somebody who gave a sample might have to wait for two weeks for the result to come in at the worst of times. And by that time, they would probably be over the infection. By the time they got actionable information, the infection was over. In India, ICMR took a different approach. They rationed the tests.
00:30:00
Speaker
They said that you can only get a test if your doctor says that you can get the test, you have to be symptomatic and so on. So that led to its own set of problems and so on. So it was clear that the information we needed was essentially to identify who is positive.
00:30:18
Speaker
and in as short a time as possible. And also if we can get it at a cost that everybody can afford, that's how you do it. And the bottleneck was that labs had to handle lots of samples.
00:30:32
Speaker
So if I have a campus of a thousand people, I just collecting thousand samples, sending it to the lab, getting them to test everything was not easy. So one machine in one go can test hundred of them. You need to run it for like 10 rounds and each round takes four hours or so. So that's like 40 hours by the time you get through it.
00:30:54
Speaker
Essentially, this is a capex problem. If you spend on buying more machines, you ramp up capacity. If you spend on buying more machines, you ramp up capacity, but you also need more space, you also need more trained people. The sad truth at the peak of COVID was that there were no more machines.
00:31:13
Speaker
So, all the extra capacity was sold out. So, you had to bypass CAPEX. You had to solve this problem, if possible, in a different way. Which test are you talking about here? There are different types of COVID tests. So, of course, my approach is, I say, independent. It's at the algorithmic layer, but we were targeting the RT-PCR test.
00:31:33
Speaker
And what is the difference between the types of tests? Like for a layman who doesn't understand? Sure. So the RT-PCR test is the gold standard test. So think about your sample like an ocean. And think of the coronavirus as some fish in the ocean.
00:31:53
Speaker
So, what the RT-PCR test will do is if there is even one fish, it will make it into two fish into four fish into eight fish and so on till the entire ocean is flooded with fish and you cannot help but find that fish. What an antigen test will do is it will put some fishing rods into the ocean and hope to catch a fish.
00:32:12
Speaker
So, it doesn't do amplification in this way. So, the PCR test uses polymerase, which is an enzyme literally given to us by God or nature, whatever you want to call it, right? Polymerase can amplify. It can make copies of DNA. It is what is responsible for the division of every cell. So, the exponential growth of life is because of polymerase.
00:32:32
Speaker
and where is polymer is derived from like it's a chemical compound you can make it in the lab. So biology today has come to the stage where we know that DNA makes RNA makes protein and so
00:32:48
Speaker
Polymerase is a kind of protein. So we say protein, we think about things in food. But when a molecular biologist hears protein, they are thinking about nanorobots. And within a cell, there are like 20,000 types of nanorobots. And each very specialized to do one thing. And each nanorobot has a family of about 50 or 3,000, 5,000 identical copies of that nanorobot.
00:33:16
Speaker
We know how to make polymerases, we can make it quite easily and we can sort of use them. Okay, got it. So, these polymerases allow copying and copying allows exponential amplification which nothing without polymerase can do. So, PCR in that way polymerase based nucleic acid tests are unique in that.
00:33:40
Speaker
Yeah, so we targeted the PCR. So polymerase chain reaction. So making 1 into 2, 2 into 4, right? That's like the chain reaction you get for nuclear fission. It's the same kind of idea. Yeah, so PCR was our target, the RTQ PCR test and the insight was
00:34:04
Speaker
RT stands for reverse transcriptase. The point is the COVID virus doesn't have DNA. PCR is good at detecting DNA. The COVID virus has RNA. So you need to do a translation from RNA to DNA and RT does that. So reverse transcriptase converts RNA to DNA before you can then amplify. And the Q stands for quantitative. So this test doesn't just tell you whether there is fish in the ocean. It tells you how many.
00:34:31
Speaker
Okay, okay, okay, okay. Got it. Concentration of fish in the ocean, yeah. So you know how severely someone is affected by COVID? Not necessarily, all the number of copies of the virus in that sample. Now, whether that is the same as the number of copies of the virus in the person is already, it depends on how much sample you have collected and whether the number of copies are correlated with severity is again another thing, right?
00:34:56
Speaker
analytical truth and clinical truth, there is a gap between the two. So we don't concern ourselves with clinical truth. That is in the domain of doctors. We are much more modest. We want to say we have in front of us a tube. Does that tube have the thing that I'm looking for? Okay. Okay.
Molecular Compression Innovation
00:35:14
Speaker
Okay. So you wanted to build an RTQ PCR test, which would not require significant capex.
00:35:22
Speaker
No, that's not how we were thinking. So we were wrapping around existing workflows, existing machines. So we didn't want to reinvent anything that was already there. What we did was we collected, we did a molecular compression. And what I mean by that is, let me give an example. Suppose there are 16 people.
00:35:44
Speaker
Okay. And now suppose I know somehow that one of these people has COVID and all other 15 don't have COVID. And my only job is to find which one person has COVID. Okay. So what I do is I arrange 16 in a four by four grid and I take everybody in the first row. And that's four people. I take samples from each of them. I mix all the samples into one tube.
00:36:11
Speaker
right similarly for the second row similarly for the third row fourth row first column second column third column fourth column so I got eight tubes and each sample has gone into two tubes right row tube and column tube so now if I see that the third row and the second column have come out positive in my tests then I know the sample in position three comma two third row second column sample is positive
00:36:39
Speaker
So, essentially we do a version of this on steroids. So, we take a thousand samples, we arrange them in a large grid and we do row column and diagonal and then we take the quantity in each, we convert it into a linear equation, we get a system of linear equations, we run some algorithms to find sparse solutions on this.
00:37:00
Speaker
And this is, so this is already, these kinds of algorithms are used very commonly in communication networks. So sometimes called LDPC codes and things like that very closely related to this. They are used in MRI machines. Previously MRI machines used to take 45 minutes for a scan. These days they are down to 15 minutes for a scan exactly because of this kind of compression algorithms running in the background.
00:37:23
Speaker
Of course, for any particular application, you need to model the underlying process very well so that this works. So you need to understand PCR very well to make this work.
00:37:36
Speaker
How does that combined test tube happen? Is that done through a robotic arm or something? That's a great question. And in fact, when we came up with the solution, we had a very hard time getting labs to actually do the mixing. It was very challenging for them. It was very complicated.
00:37:56
Speaker
So these would be like, not just milligrams, but even less than milligram quantities. Yeah. So, so they take, they take pipettes and they were manually taking liquid and then putting it using pipettes. And it was taking them several hours to do this and eating away all the gains that we were generating with the algorithm. So then we had to enter into solutioning for this. And we, so we came up with matrices. We came up with the patterns of pooling.
00:38:24
Speaker
that would be particularly easy to do in the lab, even if done manually. So we came up with kits that would make the pooling very easy to do. So all of that innovation had to happen to actually take this from an idea to a solution.
00:38:44
Speaker
So talk to me about that journey from idea to solution. So I understand the idea which you have that if you have 10,000 samples, you don't need to run 10,000 tests. Maybe you need to run what, 500 tests? What is the ratio like?
00:38:58
Speaker
So, I mean, maybe 300 tests. Wow. Okay. Okay. So, which means that you are significantly ramping up capacity, maybe by something like 300 factor of 300, you're ramping up capacity. But usually you don't get 10,000 samples in this case. So it's like more like 1000 samples, and then you do it in 100 tests. So we have done 1000 in 100. Yeah, so that I can confidently talk about.
00:39:27
Speaker
So you have a theoretical solution to increase capacity of RT-PCR testing by 10x. And at this time, you are still in IIT. So what next? You thought of this in IIT, then what? Did you get some students together? Yeah. So I have a WhatsApp group with my students and my research group.
00:39:54
Speaker
So I explained this idea on my WhatsApp group. One of my PhD students got very excited about this. And so he ran with it. He kind of wrote the first code and then he showed with his code that yeah, this all this seems to be working. And then I wanted to try this out in the lab. So I went to a collaborator, a colleague at the National Center for Biological Sciences. And so
00:40:24
Speaker
No, it's in Bangalore. So National Centre for Biological Sciences happens to be an institute within the TIFR family of institutes. So I had visited there quite often when I was at TIFR and I knew him well. So I told him about the idea. He got very excited.
00:40:41
Speaker
And then he helped me find the right person at NCBS who could help take this forward. And yeah, so what does that mean, take this forward? Like, what does that mean? Do the experiments in the lab. So they had, they had some, initially they had some synthetic DNA, synthetic RNA that they tried it with. And then later on, they had... Synthetic COVID RNA.
00:41:05
Speaker
Not even COVID. This was just some planarium or some random species lying around. The principal was tested that this would work. And yeah, separately, my brother was at Harvard, Harvard University. And so he again is in this field of DNA nanotechnology. So he's, yeah, so he's a postdoc. He's there. And so he got excited by this. And so they had a liquid handling robot. So they wanted to try it out with that robot. So they got synthetic COVID RNA and they tried it.
00:41:35
Speaker
covid DNA actually and they tried it with that and both these experiments work the first time they work quite well and so we had a strong group of principle and then in my mind again so naivety I thought it will be
00:41:50
Speaker
I thought government will run with it. I don't have to do anything more and it will be everywhere. They'll buy these liquid handling reports. Of course, nothing of that kind happened. And there was also this. How did you try to get it commercialized? Like did you submit it to somebody that you should try this or like, what did you do to get the government to run with it? So, so the, through NCBS, we sort of got in touch with principal scientific advisor to PMO.
00:42:18
Speaker
And so he was very encouraging and he sort of said that, yeah, you can definitely go ahead with this. Director of NCBS got involved and sort of helped us in some ways in what our next steps. And so then the idea was that we need to do some more experiments with patient samples and then prove this. So some of those started. And yeah, so with a hospital in Bombay, we started some of those experiments.
00:42:48
Speaker
And this was like a project under NCBS so far. This was like a project. So this was initially a project under IIT Bombay, and then NCBS came in. And so then a project under NCBS. And yeah, so, and so that happened, right? And yeah, things. So at that point, I think time was slipping very fast, right? I mean, we had built something, but
00:43:14
Speaker
Yeah, I think first wave had hit, second wave had not yet come. But yeah, I think the risk was there. And we were very keen on trying to get something before the second wave. But I mean, learnings, right? I mean, we started talking to labs as well, asking whether. So very early on, we had formed a team, a volunteer team of people who were just excited by what we were doing to call up every testing lab in the country and sort of find out what equipment they had.
00:43:40
Speaker
and whether they would be able to run the kind of approach that we were suggesting and whether it would be useful for them. And so we knew what labs had, what kind of equipment they had, what they didn't have. We knew they didn't have robots. We knew who had robots and how many. So we sort of calculated all that information individually, calling up people or looking them up on LinkedIn and then getting their number. And I think we didn't even know how to approach for
00:44:08
Speaker
regulatory approval or how to go about it. I think we were fairly lost in all of those things. We just didn't know who to ask and who will approve it, who has to approve. And was it ICMR? Is it somebody else? And I think some of those things were not clear to us. And then at some point, one, so I got in touch with Nidhi from Axelor through a student who had worked on our project. What is Axelor?
00:44:35
Speaker
So, Axilor is a venture capital fund by Ex-Infosys founders. So, Nidhi is one of their venture partners. So, she got very excited about what we were doing and the potential for this as a business beyond COVID, also to build many other things and so on. So, Nidhi helped a lot in laying down the groundwork for how to approach regulatory agencies, how to think about
00:45:05
Speaker
Yeah. So I was saying in my mind, I had already decided that I'm going to start something. And the point was when I realized this technology was multi-use that we could essentially build it once and we could deploy it to many use cases with some, of course, one would have to do solutioning. One would have to invent new things, but I saw a path. I saw a path. And when I saw the path, I kind of got excited about it and I wanted to start.
00:45:27
Speaker
So when Nidhi approached for a conversation, I was actually happy to talk. I was looking for those kinds of conversations with VCs at that point. And it also fit in with my plans to eventually go start up something on my own. It just happened earlier than I thought it would happen. The COVID effect, I suppose, some extent.
00:45:45
Speaker
So, yeah, so that's how it happened. And so the, so Axilar has come in as an investor and with their help, we are sort of taking this forward to sort of multiple other areas, including COVID testing and diagnostics is one area, but we are also looking at food testing. We are looking at in the seed research for agriculture. So yeah, so we are looking at multiple other areas, some initial conversations in pharmaceuticals and so on.
00:46:15
Speaker
So like Nidhi said, okay, I'll back this as a venture. Then what, like, did you raise a seed round? Did you like hire people, build a team? What was your go-to market? Like, I want to know that whole story of... Yeah. So we raised a seed round. We hired people, built a team, all that. So that's something I would rather not share at the moment.
00:46:38
Speaker
So, so we raised a seed round and we, like some of the hiring started even before the seed round money came in. So one person who had, who had known from my TIFR days, she came in as a product manager and she had experience with Bangalore based unicorn as a product manager and so on.
00:46:56
Speaker
and a biology base, strong base in biology. So I think that really helped having somebody strong there. I got one more person again with some years of software engineering experience from Bangalore who came in as a software architect for us. And we hired a recent PhD from TIFR as a molecular science specialist to make sure that the solutioning we were doing wouldn't interfere with the biology of what was going on.
00:47:24
Speaker
And then we hired a couple more PhDs. So one data scientist who's located in Los Angeles is on our team. So right now we are a team of some five PhDs and some five more people with industry experience and some five more people who are interning and might be graduating soon. So as we expand, some of them will also get absorbed into us. What is the product?
00:47:51
Speaker
Like you have an algorithm to improve efficiency of testing. How does this translate to a product? I mean, this could very well also be just published in a white paper and various testing organizations like say, Dr. Lal path lab, they could just adopt it because it increases their profit margins. So they have every incentive to adopt this.
00:48:15
Speaker
So what is your product? I'm not clear
Startup Development and Market Strategy
00:48:18
Speaker
on that. Yeah, yeah. So the algorithm is not enough. I mean, you can't just use the algorithm and do this because you need the solutioning, right? You need sort of the end-to-end solutioning of this.
00:48:30
Speaker
And so that's where we come in. We have built a web app, and we give a pooling kit to the lab. We give training to the lab on how to do this. And using the pooling kit, they're able to do the pooling fast. And then they export the results from their machine onto our web app. The solve happens on the cloud. They get back the results. So that's the workflow. What is the pooling kit? Just talk to me about these individual items.
00:48:57
Speaker
Yeah, so the pooling kit is so we have these matrices for pooling which decide how the samples get mixed. So everybody is familiar with test tubes, right? It's inconvenient in a lab to work with test tubes when you are going to be handling a large number of samples.
00:49:14
Speaker
So one thing that is commonly used in a molecular biology lab is called a plate. A plate is many test tubes together in a grid. So the test tubes are like wells in the plate. You call them wells. You don't call them tubes anymore. And so we have built a plate of the same dimension. But instead of having wells, that plate has lanes or troughs. And so those lanes or troughs are facilitating the mixing.
00:49:42
Speaker
Okay. But how does the, like the liquid is right at the bottom of the test tube. How is that mixing happening then? Like because. So, so the, the samples are lifted using a multi-channel pipette. So eight samples are lifted at a time, put into the plate and then another eight are lifted, put into the plate. So just by transferring the samples from the original plate to our plate, the pooling is accomplished.
00:50:10
Speaker
So you don't need to think, you just need to transfer. And once you transfer, the pulling is accomplished by the geometry of the plate being appropriately designed. Okay. Okay. So, so it has like a slant or something which enables the liquid to travel. So slant is there, but that is not the point. The main point is that which wells are connected. So the wells that are connected end up in the same pool.
00:50:37
Speaker
So there's a trough connecting various wells. So this is one way in which you productized it, creating this plate like a pooling kit. What are the other ways in which you productized it? What is the digital element of it?
00:50:55
Speaker
So, I think we have a sort of mathematical model for the underlying assay itself, PCR, which is tied very closely with how we are doing our inference. So, the algorithm requires both. It requires the modeling and it requires the inference.
00:51:11
Speaker
And the whole thing is delivered on the cloud through a web app. And we have to make sure that when people upload the data from the machine, we are able to take that data in a seamless manner and then we are able to send it to the algorithm and give the results back.
00:51:30
Speaker
So when does the lab technician use the web app and how does he use the web app? So as a lab technician, I've got a hundred samples. So first thing is I will transfer them to your plate. So you've got a thousand samples, let's say, right? And you have 10 such plates. So you transfer all these samples to different plates and then you transfer it finally into the plate that is going for PCR. So that's a hundred sample plate.
00:51:58
Speaker
and this 100 sample plate you test and the PCR machine gives you an excel sheet with the results that so you go to my web app there is an upload button and you upload this excel sheet there and then you hit solve so it does the solve there and it will tell you that from your original thousand which are the samples which have which are positive like okay okay okay so it gives a report for all the thousand samples
00:52:28
Speaker
with the, yeah, quantitation and then for every gene tested a separate quantitation. The original thousand samples data also needs to be uploaded, right? Like for you to like patient name for each of those thousand patient names, you would need that to be uploaded so that you can give. So yeah, so we have solved that problem by sort of having, so we have solved that at different stages, right? When we do the collection itself, some kind of register is maintained and we know it already at the web app side.
00:52:57
Speaker
So, the lab doesn't need to enter it. As far as the lab is concerned, it is seamless. They just get some plates, they do the pooling, they do the testing, out comes the results, they upload it and they get the thing, right? And they will have, so they get to see one sheet which has this mapping that is pre-prepared for them.
00:53:15
Speaker
that sheet is pre-prepared for them. So they can always check that the right samples are going into the right well as spools, so that they can check. And if they have made any mistake, they can directly go into that sheet and edit. And so that flexibility is there. So we have gotten rid of absolutely any manual entry from their site.
00:53:34
Speaker
But how did you get the initial data on your system, like the sample, the patient came with some unique code or something to identify the test tube?
00:53:46
Speaker
Yeah, so we have a code set up in that way, exactly. And the patient name versus this code, that map is done at the point of collection. So there is a registration sheet where those are written as the patient comes. And that doesn't come into our system, just the code comes. We give the solve in terms of the code.
00:54:06
Speaker
And so yeah, so we don't take the patient name at all. This is on purpose because we are doing a screening and not a testing. So we don't want to carry the patient information directly at this stage. What do you mean you're doing screening, not testing? What does that mean?
00:54:21
Speaker
Well, so, I mean, there are some ICMR regulations, right? So, in terms of testing and so on. So, what we are doing is we are collecting samples from everybody. We are testing everybody on campus, right? Using this kind of a pool method. Now, if we are calling this testing or a diagnostic testing, then we will have to enter, we will have to sort of comply with some ICMR
00:54:46
Speaker
Things like we will have to enter every sample into the ICMR website and so on. So instead of that, we are not collecting names. So the names of people are there with the campus. The mapping is there. We only have the anonymized code for every person and we give the results in terms of that code. So that kind of, yeah, so it provides us being seen as a testing and not a screening.
00:55:11
Speaker
Are you, so is this a service you're giving to labs where labs can subscribe to, so say, for example, as an add-on feature, like you have software, which, which you can subscribe to, which makes your other software more effective. So in a way, is this like that, like a subscription for a lab, which they can plug into their existing workflow so that they are more profitable or.
00:55:36
Speaker
Are you the ones who are actually selling this to the consumers? Is this B2B or B2C? We are going directly to campuses.
00:55:47
Speaker
and we are selling to campuses. We are using labs as our vendors. This is not the model we started with. We started with directly going to labs. Labs were very keen, very interested, but hesitant because they want the industry to accept it. Even if you have regulatory, there is a certain stage of acceptance after which a lab becomes comfortable. But when we were taking business to labs, then it was easier for us to actually offer this. So we went to campuses. Now we are actually seeing some of the labs comfortable
00:56:17
Speaker
to take us on as a subscription. So I think we are seeing that comfort built after they have done so much testing. So I think that's how it is. Okay. So initially when you started, you wanted to sell as a subscription to labs, but because that was not getting you traction, so you decided to
00:56:37
Speaker
do like a pivot to consumer play, like where you are going directly to the consumer. So the consumer is still a business, right? Still a campus. So that's still B2B. I'm not selling to an individual student. I'm still selling B2B. Yes. So we started that way. That has allowed us to give the labs comfort because they get to see our testing and the results week after week, and they get to verify themselves that things are working at our expense, right? So we are essentially paying them
00:57:06
Speaker
And so that comfort as they have gained, now they have become more comfortable in considering that subscription model. Yeah. So subscription model makes perfect sense. It's just that there was a, there was a hill to climb before that could happen. Yeah. So what do you price your subscription at for a lab?
00:57:28
Speaker
So for a lab, they are going to be paying about say some around 40 rupees or so per result. That's quite steep, no? Because I think the cost of a PCR test is about 500 to 1000.
00:57:48
Speaker
Yeah, so it's if you think about what they are saving, right? If you think about the fact that they are able to do 1000 in 100, right? And if you think about it in terms of increase in throughput, that a lab that is today testing 1000 samples can test 10,000 samples or something that's yeah, it's not much.
00:58:05
Speaker
Right, right, right. Yeah. Okay. Yeah. From that angle, it is fairly profitable because that like what a lab pays for. Yeah. Okay. The economics makes sense completely for a lab. Like they get 10x more productivity. So at just 10% of the price of a test. So that remaining 90% for that extra 9x increases pure profit for them.
00:58:30
Speaker
Got it. Okay. Okay. Okay. And so, and when you were selling it to campuses, so what is the pricing there for a test? Like, or is that like a regulated pricing? Yeah. So for campuses, it was about say 250 rupees per result. Okay. Okay. So what is the long-term play for you? Like, do you want to be a testing company or do you want to be a subscription company selling to labs?
00:58:59
Speaker
So, we see ourselves as a software company and both the models, as you are currently seeing, both the models can exist. It can be subscription, it can be customer. We believe that we are going, we will go through a journey. Every market we enter, we will go through a journey. And the first place we will enter will be between customer and lab.
00:59:22
Speaker
like we have done here. We will be between customer and lab. We will be giving the customer much lower cost than lab is giving. And we will be giving business to the lab from customers who would not otherwise come to the lab. And this would be a win-win for both. We will make that market. And in the process, we will also get learnings. We will be able to productize our solution to a level where the lab also changes their thinking. They also start to believe in this. And at that point, we will be able to flip
00:59:50
Speaker
we will be able to get the lab to adopt our solution as a standard in the industry. And then we just sell to the labs. So I think it is going to be this kind of journey. We'll have to stand between customer and lab initially. So zero to one is going to be that. Then one to 10 journey is going to be that and then try to flip. And then also growing through channel partners is very much on the cards for us. What does that mean? What is the channel partner for you?
01:00:20
Speaker
So, we don't, so we see ourselves primarily, our strength is going to be in IP, right? In generating ideas, innovations, and then creating products out of them. We don't want to become a sales engine. We want to find people who can take the product to market for us. Once we have proved the model, I think people cannot prove the model for us. We have to prove that first one.
01:00:47
Speaker
We have to take it up to 10. We have to show how the model scales and what are the economics, unit economics. Once that is established, we want to go with a channel partner who will help us take this to market and then help us scale in a big way.
01:01:02
Speaker
So give me an example of what do you mean here by a channel partner? So, so, so for example, in the space of agricultural seed research, we have been working with a company which is in the lab management software ERP software kind of space for these companies. And so they know everything in that industry.
01:01:25
Speaker
And they also understand our software. And for us, they become a sales channel. So they can go and talk to 20 people and they can find the five people who might be most interested and put us in touch with them and help us close the sale. And once the model is found, then they can help us sell it to more and more and more people. So it can even be a bundling at some point.
01:01:48
Speaker
Okay, so I'll come to this like allied use cases like the seed research one and the other allied use cases, but just staying on testing like this specifically that the medical use case of it, isn't there a
01:02:03
Speaker
like a very large business to be built purely as a B to C business in the space of testing where you are able to give quicker results to people and you are able to give them lower cost of testing and more reliable answers because you are taking like a more full stack approach of
01:02:29
Speaker
And I mean, you can build processes and workflows to deliver better quality. So isn't there a large business opportunity in that to disrupt traditional labs and make those labs as your vendors or make them as a branded lab for you? Like what say, all your rooms did for unbranded hotels. So it gave them a branding, it gave them customers, it gave them leads and it in short, consistent quality through software.
01:02:58
Speaker
So I agree. So I think that a lot of testing is going to reach home or become home collection and get disrupted in that way, even if there is a lab back end. I think that is something in play and you will see much more edge testing than central testing in the years going forward. A lot of home testing kits will take the market.
01:03:20
Speaker
What is edge testing and central testing? Well, I just made up that name, right? So like cloud, you have edge and cloud, right? So edge devices. So you're going to have edge testing where people are going to get tested in their homes. So today we have the finger prick test for diabetic sugar, right? So you're going to literally have a lot more tests like that available to do at home.
01:03:42
Speaker
So that's going to happen and we do see that happening. So for us right now, our focus is on getting to market fast and the problem with medical diagnostic space is that it can take longer to get to market.
01:03:58
Speaker
So that is that is then going to affect cash flow. So for us, our strategy, this is on the horizon. This is in the vision. We know how we want to reach there, where and when we want to reach there. But the road to reach there is often not as direct. So if we go after it directly now with current cash flows, it will not be a good strategy. So that's why we are currently focusing on industries like food and beverages where we can get to market in like literally a month or so.
01:04:26
Speaker
and we can start making revenue there. So that makes sense because then we will have enough cash to actually keep the business alive and to actually go forward. So that's how we are thinking about it.
01:04:37
Speaker
So you're saying for the medical diagnostic market to scale up there, you would need to spend significantly on, for example, brand building and customer acquisition and maybe set up a collection infrastructure. So even regulatory, right, even regulatory, you may, if you have get a regulatory clearance, you have to spend money. But you must have got regulatory clearance for doing it on campuses now.
01:05:01
Speaker
Oh, that is only one type of test. So for each test you will need to do it separately and internationally there is still a cost to that. All of that can be done. We are confident we can do that. And the challenge is doing it at the right time. So it's one single FDA approval is going to cost $3 million.
01:05:22
Speaker
So, one has to have that kind of check size and one has to be prepared to spend it just for that with a clarity in market. And so, for every company, right, when you are expanding, you want to put your foot very carefully on solid ground and expand one at a time. So, the way our vision for this market, if I think about molecular testing, it's a horizontal market and the size is about 62 billion US dollars.
01:05:46
Speaker
And what do you mean by this term horizontal market? So there are multiple verticals. So there is pharma, there is food and beverages, there is medical diagnostics, and there is industrial microbiology. So you can sort of dig into these four. And so molecular testing is used. The same kinds of technology are used across these four different verticals. In that sense, it's a horizontal. It's very consolidated. It contains labs. It contains people who make kits.
01:06:14
Speaker
It contains people who make machines. And then it contains a fourth category, which is effectiveness multipliers. So effectiveness multipliers might be like robotics or the kind of solutions we are offering. So we are starting off on that fourth bracket, which is a niche, which is 15% of this market growing at 15% CAGR, rapidly growing small niche worth about 10 billion US dollars. So we are starting within that.
01:06:38
Speaker
And within that, instead of starting in, so should we start in medical diagnostics? Should we start in pharma? Should we start in food and beverages? Or should we start in industrial microbiology? This is our question. So we have started pilots in multiple of these areas, but there is the natural speed at which the pilot is going to translate into commercialization. And that's going to happen fastest in food and beverages. So that's where we will end up offering the first post COVID commercial products and growing initially.
01:07:07
Speaker
It will happen naturally that the other cases will catch up and we will be able to productize in some way. We will be able to have offerings if we want to go to home use or if we want to go to that kind of use tomorrow, whether we choose to do that, whether it fits into the business at that point, whether it makes business sense, that is still a little far in the future. So, but yes, this transition that you're observing is very well observed and it is true that medical diagnostics is going to move in a direction that is disruptive. Yeah.
01:07:37
Speaker
Okay. Okay. So now I understand molecular testing use case for medical diagnostics. Obviously blood testing is molecular testing. Help me understand use case of molecular testing in the other three like pharma, FNB and industrial biology. Sure. So in the FNB sector, you can have all kinds of commodities. So tea, coffee, spices. And when these are being grown, pesticides are being sprayed.
01:08:07
Speaker
and certain level of pesticide is allowed by various food authorities.
Industry Applications of Molecular Testing
01:08:13
Speaker
Now, if you are going to export these commodities and you exceed this level and your product gets tested and found to be exceeding the level, then that has cost all the product recalled from the shelves with punitive measures. Some countries have very strong punitive measures which can include lifetime bans. So this is a big risk for that sector.
01:08:37
Speaker
And they are trying to get away with some cheap kind of tests and so on. So some of the tests which are like paper strip tests, A, they are not sensitive enough. So they can't quantify small amounts that are still in violation. And B, they can test for like one pesticide.
01:08:54
Speaker
but the number of banned pesticides is 500. So all that it will do, even if it is effective, is it will move the use of pesticide from pesticide A to pesticide B. There are gold standard tests that can test for all 500 at the sensitivity required, but they are too expensive. So you can make those gold standard tests affordable, basically. We can make the market meet.
01:09:19
Speaker
So that's, that's what we are, we are the, yeah. So that's what we are doing there. And so that's the food and beverages play. That's one dimension of it. Right. But of course. Then you are directly doing subscription model. You're not selling to the, for example, the, the actual packaging companies which are doing the packaging, you're not selling to them, but you're selling to the labs who do the testing for the packaging companies through the channel partner. No, no, no, no, no, we are not selling to labs. We are selling to the packaging companies.
01:09:49
Speaker
Yeah, because they are the ones who have the need for that service. Labs don't have big business from them right now. The market doesn't exist. They are just not testing. They are taking the risk. And they are getting hit by all of this. So that's how it is. And then they are trying to work on trust. But one bad apple spoils the whole ones. So it's a big risk. They are not able to actually control this very well in this industry. So I think that's the play.
01:10:16
Speaker
Okay, so you're going to packaging companies and saying that I will run the gold test for you at maybe one fourth of what it would have typically costed you. Exactly, exactly. So we are doing a pilot with India's largest ice exporter, the Eastern Group. And so they are taking this very seriously. And so that it's a paid pilot we are doing with them. It's we are hoping that yeah, it can. And again, the lab is like a back end partner to you. You are currently Yeah, yeah.
01:10:47
Speaker
Yeah, so I think once the value of this gets established and a lot of business is being generated, then there might be other possibilities in terms of the model and so on. A subscription part of it. If one wants to go to it at that point, yeah. Again, same thought occurs to me here that it makes more sense to continue to build relationships with packagers than sell subscription to labs because you can ensure quality in a better way than a lab would be able to through digital tools and
01:11:16
Speaker
Yeah. So we have thought about this. We have some thoughts in this direction and we have some thoughts on how the product roadmap should unfold. And that also ties in with this. Yeah. So, I mean, these things will play out. You will see the moves as we make them. Okay. And a pharma.
01:11:33
Speaker
use case in pharma. So so so pharma is still very initial what we have learned are some tantalizing bits here and there where we think we can contribute but no pilot on the table yet we are still in initial stages so looks like for clinical trials pharma would like to know if it is working for patient A not working for patient B what is the difference right and they would like to monitor patients so that if a clinical trial is failing they know on day three rather than day 17 right by the monitoring of the patient or
01:12:02
Speaker
even animal model and all kinds of other use cases in terms of drug discovery, searching for a large searching for a particular drug in a large collection of molecules. So again, we are good at those kind of searches and we might have a role to play. So yeah, still I think initial days with pharma. The other place where we have a pilot ongoing is with Mahiko, which is India's largest seed research company.
01:12:29
Speaker
And so the way that this works is a seed research company is selling seeds. They have a very active market research group who is looking at how these things are performing in the field and what the customer wants differently from what is available, takes this market info back and they decide on a product.
01:12:50
Speaker
and a seed with certain properties. Then the breeders start trying to produce that kind of seed, and that can take like a million experiments and eight years of research. So by the time that seed is in the market, climate conditions have changed, pest patterns have changed, lots of things have changed. So that eight years, if we can cut it down, that's a big game changer.
01:13:11
Speaker
How will you cut it? Yeah, so right now a lot of their effort, a lot of their time is going in, a lot of molecular testing. So if we can turn it around faster, if we can accelerate that molecular testing, then they might be able to increase the numbers that they are trying in every batch. So instead of trying 100, they go to 1000, instead of 1000, they go to 10,000.
01:13:33
Speaker
And so if they are searching faster in every round, they may get to the final destination faster. So it seems to me that your playbook, like a slightly long term playbook is, and I mean, correct me if I'm wrong, I'm making an assumption here, but it is to find testing use cases and focus more on
01:13:54
Speaker
businesses who have testing use cases. So maybe creating a Dr. Lal path lab would not be appealing to you, which is like a consumer testing business, but something which is a business testing business where like say,
01:14:10
Speaker
people who are packaging FNB products or companies which have like who are running clinical trials and need testing or like seed research institutes who need testing. So like a B2B molecular testing business is what you want to be in an asset light model. You would not want to actually build your own labs, but you would want to be the layer of software which is
01:14:34
Speaker
making the whole thing work cheaper, better, better customer experience and much lower costs through the layer of software. What's the roadmap to achieve this vision? What do you see as the steps that you need to take to get there? There is the kind of natural unfolding in terms of commercializing and then
01:14:57
Speaker
finding growing within that vertical and then growing across geographies and then finding new markets to commercialize, finding the right channel partners again growing and across verticals and across geographies. So I think that will anyway unfold. And I think one important thing is also having this things are going to change right faster than they've ever changed before.
01:15:19
Speaker
and molecular testing is going to be disrupted and it will be very different three years from now than it is today. The kind of technologies available will change and there will be a convergence. Right now we are thinking of automation, we are thinking of chemistry, we are thinking of algorithms as different things as competitors, but they are all going to come together on single platforms
01:15:42
Speaker
and you're going to have lots of interesting things happening there. So we have some thoughts as to we have our own bets on how this is going to play out, where we need to position ourselves and so on. So some of that product development, R&D, spending. But tell me about those thoughts, those bets which you want to take.
01:16:02
Speaker
Yeah. So, so, so what is the vision, right? So, so the vision is that there are three access for technology. So there is cost, speed and power of the test. So ideally you would like to make a test more powerful, faster and better, more affordable, more affordable, more powerful and faster.
01:16:24
Speaker
and that's the direction in which it is going and how do you so usually there is a trade-off if you can get two of the axes you are losing out on the third. What algorithms is allowing you to do is it's allowing you to make the triangle itself bigger.
01:16:39
Speaker
And I think that is a power that we want to utilize in many different situations. So people who care about cost, that is one play, but people who care about accuracy, we can improve accuracy keeping the other two the same.
01:16:54
Speaker
And so I think making that triangle bigger is really what is exciting for us. And how to make that triangle bigger in situations where there is a good niche for us that is really the key question, right? Where that kind of gain cannot be competed with any other technology that is especially interesting to us. Okay. The three like cost, power and what was the third, sorry? So cost, speed and power. Cost, speed and power, right, okay.
01:17:25
Speaker
So power is not something which you are per se getting into, right? Because you're not going to think of better tests, but you're going to think of tests which are already there and make them affordable by the algorithm, right?
01:17:40
Speaker
Yeah, so no, I think we are going to get into better tests, but we are going to be able to get into better tests in an off-the-shelf manner, where what we are contributing is the new algorithm, but what is the end result is a better test.
01:17:56
Speaker
So let me maybe try to explain this point a little bit. So currently, as I said, the algorithm can make the triangle larger. So you can take tests that are not great on power today, but are very affordable and very fast. And using the using these algorithms, you can make them more powerful. That's also possible. And we have some ideas currently too early to talk about, but in the lab right now.
01:18:24
Speaker
that we are working on, where we can take it in this way. And putting it together in this way is, how do I say this? How would you make a test more powerful using algorithms? So if you think about it, what's a good example? So you think of your hard disk. There is information on your hard disk that you're reading. Now the individual bit on your hard disk need not be very reliable.
01:18:53
Speaker
because there are error correction schemes. When you read an entire sector, then there is some parity check code for it. So, individual bits can be wrong. Similarly, you can do similar, I leave it at that analogy. So, you can actually make a more powerful test
01:19:11
Speaker
when the physical layer, the chemical layer is not so great. So things like this can be done algorithmically. So by reading more molecules, you can make the test better, essentially. In a way. So we have some thoughts on this, which is too early to kind of get into in any kind of detail. But yeah, maybe a year from now or so, we'll be
01:19:33
Speaker
more comfortable talking about this. But the idea is there that we believe that algorithms can improve the power of molecular and one big problem today with molecular tests is
01:19:46
Speaker
that you don't know whether they are correct. So one doesn't want to be in that world. One wants to be in a world where you can trust it and even better if you can have some kind of proof for it. So you are the kind of person who would think more solutions, but who looks after customer acquisition?
01:20:07
Speaker
Now, because I mean to really grow big, you also need that to onboard new businesses who need testing solutions at scale. Like who's looking after that? Yeah. So as I said, right, so right now, so I have a, so my CBO, Nidhi and I, we are both very hands on.
01:20:26
Speaker
in customer acquisition as well. So, Nidhi has come in as our chief business officer and that was part of the investment. So, that is the model that Axilar brings to startups that they also come in.
01:20:41
Speaker
So she has helped a lot in kind of bootstrapping the business. I have been very hands-on in the customer acquisition. I think that is not going to change. I think that will remain. Because zero to one, I think we
Scaling with Channel Partners
01:20:53
Speaker
will have to do. The product team will have to do. One to ten that scaling a CBO kind of person can take. Beyond ten, we don't want to do it in-house. We want to take it to a channel partner.
01:21:03
Speaker
So we want somebody else to grow that. So that is, once we have set the model, we think that's somebody who can come in and then, right, who's already has the market sense and already doing that. I think the right channel partner is how we are looking to expand. Yeah, I'm enjoying the customer side of things. So I'm still learning, of course, but it's I think sales is something that I'm really enjoying because what I realized more and more is that it's about listening very carefully. And it's a lot like getting a paper published.
01:21:32
Speaker
in a peer reviewed journal because when they say no, especially in a B2B setting, what's really being said is some more ideas about why it's not working for them. And if you can address those points, then often there is value to be added and then deals can close.
01:21:50
Speaker
So I think a lot of patience is necessary. I think it's a long kind of play and one has to kind of figure out exactly what model, how long things will take and so on. But so far, I think the people side of things has been something I'm enjoying. I don't understand why you want to go with channel partners. Traditionally, FMCG companies use channel partners because
01:22:13
Speaker
they want to reach lakhs of, not even lakhs, but of consumers and across a very large area. And so it is not possible for them to build the supply chain of reaching to those consumers. But you probably would have maybe 100 companies in each of those verticals which you would want to reach out to. So why not do that in-house only? Like maybe there would be 100 packagers who would have enough volumes to be worth your while.
01:22:42
Speaker
So the right word would be leverage. So I mean, I want to be able to move a heavy weight by applying a small amount of force. And so channel partners allow us to kind of become bigger without increasing our burn at the same rate. And so I think in terms of cash flow at early stages, it works better and it allows us to keep our focus.
01:23:05
Speaker
on being a business that is driven by technology that is focused on building product because we believe that's where the future is. So I've worked on molecular computing and on making chemical reactions intelligent. I want to bring some of those technologies to molecular testing. That's really the long road map. Right. And so I can try to try to save on that the channel partners and try to grow myself. But then all those salespeople will be in house. We'll have to manage. We'll have to do all that.
01:23:33
Speaker
It will become a different kind of company. We'd rather be a technology company and early on for sales we'll take help from people who know the market. Once we are at a particular scale, if things are going so well, we want to start verticalizing, it'll make sense at that point. But I think early on it makes a lot of sense. So what is the channel partner here? Is it somebody who's
01:23:55
Speaker
doing an introduction and getting you in the door, but the customer knows that it's your test and the branding is yours. Or is the channel partner someone who says, I have a new service to offer you and that service is sold. So it depends on the stage, right? In zero to one or one to 10, it is the first one, but once we have 10 customers, then it can be an integrated product, right? So if the channel partner is somebody who already has a customer for whom our product has value.
01:24:25
Speaker
And so the channel partner's product and our product can possibly be bundled together and sold in one shot. And now it's an additional value that's already going to the customer. But you're okay with your brand not getting exposed to the customer. Like the brand with the customer, he sees this as a service from his channel partner.
01:24:45
Speaker
Yeah, that's a great question. I think some of these things we will have to look at in a case-by-case basis. We'll have to think about on a strategy basis. I think at this stage, that brand consciousness for us is not the key thing. We are happy to be white label so long as the service is happening and we are able to deliver value.
01:25:05
Speaker
So you started in the US with molecular computing when you joined your PhD.
Molecular Computing in Testing
01:25:11
Speaker
So how does the connecting the dots happen for you at a personal level?
01:25:18
Speaker
Right, right. So the vision in molecular computing has always been that we can build smart systems of molecules, right? And molecular testing is, I believe, a good beachhead market for some of those ideas, where we can build smart molecular systems, which will do testing better than before. So I do see what I'm doing as a natural extension of those ideas, and I do see those ideas coming in in a bigger way and taking this field forward.
01:25:49
Speaker
Give me an example of molecular computing leading to better molecular testing. Right. So if you even take what we did with tapestry for COVID testing, which was our campus screening solution, we were treating the mixing of samples as an addition operation so that what we were doing was solving a bunch of linear equations. So again, that there is a molecular computing flavor to what we were doing there. And in some ideas that we've been
01:26:18
Speaker
exploring in one of our pilots. We have been constructing other molecular primitives like subtraction.
01:26:25
Speaker
and logic gates to be able to extend some of our coding ideas to more use cases. How would subtraction work? Like there would be some chemical action on it, which would allow you to remove something from it. So you can try that. It turns out that doesn't work very well. In fact, to do subtraction right turns out to be somewhat nuanced, but there is a well-known algorithm within the molecular
01:26:53
Speaker
computing community called approximate majority which does a great job of doing some kind of subtraction so we are looking at bringing those kind of ideas to molecular testing. And subtraction helps you in increasing power because like you gave me the example of fish in the ocean if you are able to increase the fish that's one way to make it powerful or another way to make it powerful is to reduce the water.
01:27:19
Speaker
Yeah, but you may be asking different kinds of questions, right? You may be asking whether blue fish are more than red fish, right? So instead of counting both blue and red, if you could have some way of getting the difference to talk to you in some way, then that could be more powerful.
01:27:36
Speaker
Yeah. So, so, so in, so this would be known as source coding within the coding community. So how do you, so if I'm sending a photograph to you, right, I can compress the photograph before sending it. So similarly, I, there is molecular information, I can compress it before testing. Right. But that compression needs to, needs algorithms. Those algorithms have to happen in the pot and the way they'll happen in the pot is with molecular
Podcast Promotion and Conclusion
01:28:02
Speaker
computers.
01:28:02
Speaker
If you like the Found A Thesis podcast, then do check out our other shows on subjects like marketing, technology, career advice, books and drama. Visit the podium.in for a complete list of all our shows.
01:28:23
Speaker
Before we end the episode, I want to share a bit about my journey as a podcaster. I started podcasting in 2020 and in the last two years, I've had the opportunity to interview more than 250 founders who are shaping India's future across sectors.
01:28:39
Speaker
If you also want to speak to the best minds in your field and build an enviable network, then you must consider becoming a podcaster. And the first step to becoming a podcaster starts with Zencaster, which takes care of all the nuts and bolts of podcasting, from remote recording to editing to distribution and finally monetization.
01:29:00
Speaker
If you are planning to check out the platform, then please show your support for the founder thesis podcast by using this link zen.ai founder thesis. That's zen.ai founder thesis.