Using AI Technology to Generate Coronavirus Antibodies with Andrew Satz

By: Arooba Ahmed (CC '23)

---

Cover illustration by: Arooba Ahmed (CC '23)

---

Andrew Satz is the Co-founder and CEO of technology company EVQLV and an alumnus of the Columbia School of General Studies (GS’15).

---

Could you describe what your company does?

We are a technology company that uses AI based methods to discover novel biologic drugs, specifically biologic therapies. Basically, we use AI to discover, design and develop antibodies. 


​
How does the algorithm that your technology uses work?

We model evolution on the computer. The way antibodies form in the body (whether it is in a human, mouse, chicken, or llama) is an evolutionary process that has been generated over millennia. We model that process and so we are able to mutate a starting antibody that we take from an existing drug, a drug coming out of a normal discovery process, or even computationally generated antibodies or cells that can bind to viral, ontological or autoimmune targets. Then we can make a bunch of mutants of it. Those mutants will have different characteristics which we try to optimize for the best types of characteristics to make a drug that works. 

At what stage is the coronavirus research in your company? 

We are in the laboratory right now, and we have designed our own antibodies in house. We have optimized them computationally and sent them to the lab to be made and tested, and we anticipate having those results in the next few weeks. Based upon that, we will move on to laboratory testing in humans and animals. At that point hopefully the drugs will be ready to go into what is called “IND filing,” which stands for Investigational New Drug filing. That is the document that you file to get permission from the Food and Drug Administration to conduct human studies. If I were to guess, we are about 5-6 months away from doing human studies. We are sort of looking broadly at the antibodies themselves to target different mechanisms that the coronavirus utilizes to both infect and procreate. The coronavirus has a lifecycle, so we have to consider where in the life cycle we want to impact it. 



Do these studies depend on lab research? 

For the coronavirus, it does heavily rely on the research on the structure of both the coronavirus and its target in the body. It also relies on huge amounts of data. We are relying on the order of about a billion sequences, within which we use selective amounts of data to train our models. And that changes from antibody to antibody and biologic to biologic. We also are actually not just working on coronavirus projects but also oncology projects and botanical medicines. 



How does computational testing and development of the antibodies make the process of drug development easier?
​

We still require all the processes that go into laboratory studies, it's just the discovery process that is different. Right now the discovery process takes about four to five years on average and costs about half a billion dollars. The reason why it costs so much is because of the high failure rates. Think about all the animals and laboratory equipment that is used. We are using AI to recreate that process with the target of bringing the time and costs down to less than a year and less than 10 million dollars. 

This is a significant drop, where it is reduced to less than a quarter of the time. But we still have to go into the labs and make the antibodies as opposed to just designing them, which is nuanced. Once you make them you would still have to do the same tests as when you discover them. But the way we work is that we try to get rid of the ones which will have higher chance of failure downstream. ​
​
Drug development has many steps, like discovery, optimization, looking to see whether it's safe, conducting human studies and looking into efficacy. Any one of these could lead to failure so when we are designing, we try to think about all of these parts and not just the very beginning. We still have to go into the laboratory and make it optimizable using animal and human studies. But the idea is that we are trying to fail in the computer as much as possible so we fail less in the laboratory and make the process faster overall.



Has the closing of labs and social distancing impacted the efficiency of your company? 

It hasn’t impacted us at all really. If anything it has made us busier because lots of companies and organizations can’t operate in the lab now. We are starting to see some changes as they are reopening. Lots of labs that were open for a short period of time were the ones focused on coronavirus research. 



How has the recent drop in regulatory restrictions impacted drug development?

Usually this part of the process takes about 4-5 years. I think the drop in the regulatory restrictions has had a positive impact on drug development. But then you also have the other issues where if you move too fast in science, especially in medicine, you can really do a lot of damage. You’ve seen a lot of news about diagnostic tests failing or drugs that are being tested that are killing people. We are accelerating and jumping over steps to get to things faster. I get why they are doing that, no criticism there, but it goes to show that science is hard. If we can find ways to do science in the confines of scientific method but at an accelerated rate we may be able to move faster but also move smartly. I think right now we are letting off all the rails. For good reason, but there will be unintended consequences. Take for example the news about the patients in Korea who were re-infected. Was that real, or was it a failed diagnostic test? If that is a re-infection, that's a pretty scary thing. 

You don’t want a coronavirus test that’s 60% effective, but 99% accurate. The standard for the FDA is usually really high but they dropped that to get as many tests out there as possible.



With the turn towards technology in this time, have you seen a change in the role of data science in research related to the coronavirus (and science in general)?

I think that the pharmaceutical industry has been doing data science for a really long time. Actually, I think that the first major data visualizations were in the 1700s, around epidemiology, centered around trying to figure out where Cholera in London came from. If you think of epidemiology as part of the medical industry in terms of determining drugs and infectious diseases, data science has been impacting it for a really long time. Especially for biostatisticians, or people who design clinical trials. I think that what we are seeing now is more utilization of machine learning in this space. I think it has to do with the fact that machine learning and data science onto itself has come to a forefront in a way that it had not until about 15 years ago. Mostly because of our computational capabilities. And a couple weeks ago, a major pharmaceutical company got a new head of research who is probably the best computational biologist out there. So we are starting to see a huge shift in the utilization of data science and I think it's going to keep growing as companies start to realize not only the value of the data they have but the value of extracting knowledge and action from the data itself.
​