Inside Perspectives: Industry Pioneer Vijay Pande Lays Groundwork for Applying AI to Medicine

Innovation That Matters

Vijay Pande, Ph.D., says start-up companies that want to be successful in applying artificial intelligence (AI) and machine learning to medicine and health care need investors and executives whose personal experience straddles two worlds  – biology and computer science.

 As one of the pioneers in building this new AI platform in biology, Pande followed his own advice. In 2015, he joined the tech-focused venture capital firm Andreessen Horowitz (aka a16z), from Stanford University, where he was the Henry Dreyfus Professor of Chemistry and professor of structural biology and computer science. He now leads Andreessen Horowitz’s investments in companies at the cross section of biology, health care, and computer science.

One of the ultimate benefits of AI may lie in making sense of the enormous amount of human biological data being generated not only in the collections of public and private genomic data bases, but also everywhere else, such as hospitals, doctors’ offices and individual wearable health monitoring devices.

 “One of the biggest things we are seeing right now – and I think biologists have recognized this for decades, maybe centuries –is that biology is very complicated,” Pande says. “Physics has this aesthetic of being clean and elegant, and biology is complicated, messy, and organic. But there’s this unfortunate corollary that biology may just be so complicated, the human brain couldn’t understand all of it.”

Pande began trying to unravel this complexity at Stanford, where he led a team of researchers in breakthroughs, resulting in more than 200 publications, two patents and two novel drug treatments. He remains an adviser to the scientists working in the Pande Lab at Stanford.

As an entrepreneur, Pande is the founder of the Folding@Home Distributed Computing Project for disease research, which pushes the boundaries of the development and application of computer science techniques into biology and medicine. He also co-founded Globavir Biosciences, where he translated his research advances at Stanford and Folding@Home into a successful startup, discovering cures for Dengue fever and Ebola.

At Andreessen Horowitz, Pande serves on the boards of Apeel, BioAge Labs, Freenome, Omada Health, Patient Ping, and Rigetti Computing. He also leads the firm’s investments in Benchling, Cardiogram, SolveBio, TL Biolabs, TwoXAR,, and uBiome.

WuXi AppTec Communications’ interview with Pande is part of an exclusive series spotlighting the inside perspectives of thought leaders on topics shaping the future of new medicines.

WuXi: How did the field of artificial intelligence (AI) develop in its application to biopharmaceutical research?

Vijay Pande: There have been huge strides in AI in many different fields. One of the most notable ones is image recognition. In image recognition AI can do a better job of identifying what’s in a scene than even a human being, where the gold standard is what a consensus of people would say. And this isn’t just like saying is there a cat or dog in the picture. This has been demonstrated in the medical field, such as in the area of dermatology, where AI can do better than any given dermatologist, which is actually quite impressive and intriguing. What AI is representing in dermatology is that it is training to do the work of many skilled dermatologists.

Due to these exiting areas, such as image recognition and self-driving cars outside of pharma, people have been wondering what could be done with AI in health care. The central challenge is how can you represent biopharma data in a way that is amenable? So certain types of data sets work very well, like genomics. It is almost like a one dimensional picture and it carries over really beautifully in areas such as diagnostic tests. One of our portfolio companies, Freenome, is a great example. They can use genomics plus AI to identify early stage cancers in the blood, something you just couldn’t do before.

There’s a greater challenge about how you represent small molecule drugs in a way that AI can learn and that’s an ongoing area of our research. And let me make one last point to highlight AI representation itself. Ask anyone on the street for two 10 digit numbers. That’s a pretty straight forward thing to do. Anyone who is mathematically inclined can do that trivially, correct? If you took the same 10 digit numbers and wrote them in Roman numerals and try to add them, I think it would almost be impossible. The only way any of us could do that is convert the Roman numerals into Arabic numerals, and for a really long number that would be really hard. But we do the conversion, do the addition in Arabic numerals, and convert back to Roman numerals. The point of that analogy is that with the right representation the operations are trivial and clear and with a poor representation they are virtually impossible. One of the greatest challenges right now for pharma in AI is how you represent small molecule drugs, but with advances in this area, we’re starting to see some traction.

WuXi: How are you applying research from your Stanford University laboratory?

Vijay Pande: We’ve been pioneering areas for applying these AI methods to small molecules for drug design and for other areas related to cheminformatics for things that are not necessarily medical, such as designing solar cells. For us, I think our ability to do AI allows us to do things people thought would be impossible. One of my favorite examples is that most machine learning methods work very poorly with small amounts of training data and some of the new AI methods work well even with two bits of data, one positive and one negative. With these breakthroughs, we’re also seeing uptake of these methods in pharma.

WuXi:  Is AI more of a university phenomenon than previous biotechnologies?

Vijay Pande: Arguably a lot of previous biotechnologies do emanate from universities and AI is very much in line with that. Universities and typically start-up companies are interested in this. Maybe one aberration in that trend is companies like Google getting involved as well. I think they’re interested in AI very broadly. They see health care as a place to have a huge impact and there are deep pockets. I think what is different now is that more horizontally AI companies will come into this space and they will come in before pharma. Although I think pharma is starting to realize the potential and is getting very interested.

WuXi: How would you compare venture capital interest in AI with previous biotechnology innovations?

Vijay Pande: One of the big trends we are seeing that’s different from the start-up side, is that in traditional biotech, from the early stages, the companies would be run very differently. They would require a lot of capital. Typically an investor would own a very large part of the company, 50% or more, and would help shape and build the company. That is in strong opposition to the way a tech company is built. Tech companies are capital efficient with investors normally holding a minority stake.

One of the major trends that we are seeing that is different now is because these new hybrid tech-bio companies (different than traditional biotech) are essentially tech companies – some software driven, computer driven and engineering driven rather than discovery driven – they are being built like tech companies and funded by tech VCs, and that’s a pretty fundamental difference. With that said, I wouldn’t say that these are purely tech companies.

If a pure AI company tries to get into this space without understanding drug design or biology, I think they will be real limited in the impact they can have. So I would argue that what we are seeing emerge is really a new type of biotech where it is a company that is built like tech companies but in the domain of biology and health care, and where the founders really have to be able to straddle both, and have experience in both. The VCs also have to have expertise in both. One of the things I have seen in the two-plus years we’ve been building up this fund, a16z, is many others come into this space, and the ecosystem is becoming quite mature and hybrid.

WuXi: What kind of fund do you have at Andreessen Horowitz?

Vijay Pande: Marc Andreessen and Ben Horowitz founded the fund eight years ago. They wanted to create a firm they wish they had where they could provide tech savvy entrepreneurs with networks of executive talent, tech talent, help with policy and regulation, marketing, with market development and corporate development. As a firm we just don’t have general partners, we have operating teams in all those areas.

Biotech was this place where, due to the fact of a lack of capital efficiency, the firm really wasn’t interested in participating until we saw the need for tech companies in the biotech space. I came on board in September 2014 as a consultant and full-time in September 2015 as a general partner to lead the biotech fund and support companies in this space.

WuXi: How will AI change drug discovery and development?

Vijay Pande: In all stages of drug development, AI can have an impact in each one of them. It can help with product identification and lead development. Understanding the biology is one of the biggest challenges in drug design. So being able to identify targets is huge.

One of the biggest things we are seeing right now, and I think biologists have recognized this for decades, maybe centuries, is that biology is very complicated. Physics has this aesthetic of being clean and elegant and biology is complicated, messy and organic. But there’s this unfortunate corollary that biology may just be so complicated the human brain couldn’t understand all of it; and that the limitations we see in understanding biology translate into the complications we see in late-stage clinical trials.

So one of the earliest places we’re focusing on is the pipeline for drug design. Biology may be too complicated for the human brain, but may not be too complicated for certain types of AI; where AI can integrate the data in a way the human brain can’t, and then be able to guide researchers into interesting new places to look and new biological targets to choose. That’s one very natural place.

What we’re seeing already, and something I’ve done with my research at Stanford, is that with machine learning and artificial intelligence you can go very quickly from one hit to getting through preclinical trials, and by quickly, I mean that process could take a year or less. That greatly speeds up that process at all different stages, in terms of picking compounds and helping in synthesis, if you go the synthetic route.

But finally I think the biggest challenge, the biggest win, will be in the area of clinical trials. If you could use machine learning and AI to improve the success in Phase 1 trials through Phase 3, that would be very intriguing. You could use all the data you learn from preclinical, from in vitro and in vivo studies, to develop a better Phase 1 trial. You could use information from Phase 1 to learn about the biology in humans so you can do better in Phase 2 and so on. I think there’s a potential for AI to really affect all of these different areas and there are researchers going after each of these areas right now.

WuXi: Do you think AI will have an effect on the cost of drugs?

Vijay Pande: AI will have an impact on cost in a couple different ways. One is greatly shortening the amount of time for development, compressing the time for development will be huge. I think you were alluding that there is only a certain length for a patent window and if you can get these drugs out faster you can have more time to amortize the cost over more years. That’s particularly appealing.

The second place where it could make a difference is how much of the cost is in failed trials. That has to be paid for somehow, and so if you can have fewer failed trials, of course, that would decrease the cost. So I think those are two easy wins.

I think that where this gets even more interesting is in areas where machine learning could have an impact in ways people wouldn’t have expected. One area we are seeing is the rise of digital therapeutics. These are software programs for behavioral therapies, typically some version of cognitive behavioral therapy. And cognitive behavioral approaches can even affect disease areas, such as Type 2 diabetes.

The U.S. Centers for Disease Control and Prevention has this diabetes prevention program where a behavioral therapy approach exceeds the efficacy of metformin. Omada Health, one of our portfolio companies, has taken the lead in expanding on this program.

What’s intriguing about digital therapies is that because there is no inherent toxicity you can constantly improve the approach with data. Omada can constantly improve efficacy and build a much better product because there is a really great virtuous cycle that happens. Omada builds a better product, they get more customers, they get more data and as they get more data they can build a better product. Often we refer to this as a data network effect.

Omada is doing this for Type 2 diabetes, and we will soon see this for anxiety, depression, post-traumatic stress disorder; even something like Alzheimer’s disease. In all these neurological areas the biology is extremely complicated and these digital therapies are having an impact. So that’s a place where you could greatly bring down the cost.

Another area is an interesting twist on this. You could combine a digital therapeutic and an existing small molecule. Pear Therapeutics recently received FDA approval for this. And this combination gets especially interesting in that just as you have small molecule cocktails, this is a cocktail between a traditional small molecule and a digital therapeutic. You can take even a generic and make it best in class with low cost.

AI will have an impact in traditional drug design and the biggest impact will be in thinking outside the box.

WuXi: How do you see the AI field evolving over the next five to 10 years?

Vijay Pande: What we’re starting to see is that pharma is getting very interested in this. This is not technology they have in-house. But pharma does a great job with M&A (mergers and acquisitions), and so we’ll first start to see these technologies in start-ups and then in pharma.

As with any new technology in the bio space, it takes a little while to get adopted. Usually it works for five or 10 years before it’s totally adopted because when you’re betting your project, your career, on a given technology, you want to make sure it works. There is a point where it starts working; and that’s what we and others are proving now, that it is working.

The next stage is getting adopted. So the next 10 years will be about adoption of this by start-ups and start-ups becoming successful and pushing things and then there will be this inflection point where everyone will want to have this technology. I think that will happen over the next 10 years, and you will start to see the impact of this much more broadly.

WuXi: How soon will patients and health systems see the benefits of this technology?

Vijay Pande: I think the earliest wins will be in diagnostics. Diagnostics allow you to work within the system. We’ve already seem companies like Freenome in cancer detection and Cardiogram with wearables for heart disease. They are out of the gate and moving well.

The therapeutics, obviously, it takes a while for things to go all the way through the cycle so it could take 10 years before AI has a huge impact on that. But with that said, I think AI is just the most modern version of machine learning. In my mind I use the term AI in computers to describe the key features in a human being, because machine learning is using human derived features.

Machine learning is not going to be something that goes from black and white to color. It will gradually become more and more important over the next 10 years.

WuXi: Do you see any limitations or drawbacks in using AI in biopharmaceutical R&D and clinical research?

Vijay Pande: The only drawback is if people go into it naively. It’s not going to be a panacea and it cannot work without high quality data. There are many caveats one has to be careful about. Often when people come into a new technology they fall into two groups. One group is completely skeptical and thinks there is nothing there, and the other thinks this will solve everything. Generally, neither is correct. In the hands of skilled practitioners, this technology can be quite powerful.

There are several things that one has to be careful of, such as the quality of data, the quantity of data, the right data, and I could on and on. That would be the only caveat I would put on things; that obviously one has to understand where it works and where it doesn’t, and how to make it work.

WuXi: What are other “out-of-the-box” applications of AI?

Vijay Pande: We are starting to see AI affect new ways of thinking about health care. So a company like BioAge Labs is using machine learning and AI for longevity. I think this will have a huge, interesting impact; where now we’re not talking about necessarily having to cure every indication, but just to expand life span in general by 50%. That would delay Alzheimer’s by 40 years, instead of six to nine months, which is what typical Alzheimer’s drugs do.

It’s only due to recent advances in biology that we’re going to see these dramatic changes. It could take 10 to 20 years, but it’s going to be an exciting time ahead of us. AI could help identify drugs that would extend our life span from 80 years to 120 years. This wouldn’t mean that once you get to 80 you would remain old for 40 years. It would mean that at 120, you would have the quality of life you have at 80, and at 60 you would have the quality of life you had at 40. That’s what people talk about when they talk about life span extension. The word transformative seems too weak of a term to use to describe this. It would be a radical reshaping of health care.

Subscribe to WXPress

Receive our newsletter and information on upcoming events.