Verified by 
Artificial Intelligence
How a Pioneering AI Biotech Startup Is Revolutionizing Drug Discovery
An interview with Insilico Medicine CEO Alex Zhavoronkov.
Posted June 20, 2020
Molecular chemistry is inherently complex, and drug discovery is an arduous undertaking. Can artificial intelligence (AI) help improve the human condition by revolutionizing drug discovery and development?
We live in a remarkable time where AI is making a significant impact in nearly every industry. Insilico Medicine is a startup to watch with its AI deep learning for drug discovery, biomarker development, and aging research. It has over 150 academic and industry collaborations worldwide. Recently, Insilico Medicine added programs to its portfolio to accelerate drug discovery to treat the SARS-CoV-2 coronavirus and fight the ongoing COVID-19 global pandemic.
Alex Zhavoronkov, Ph.D., a trailblazing scientist and visionary entrepreneur, founded Insilico Medicine, an AI biotechnology company based at the Emerging Technology Centers at Johns Hopkins University, in 2014. Today, the company is based in Hong Kong and operates globally with operations in multiple countries. Investors of Insilico Medicine include heavyweights such as BOLD Capital Partners (a venture capital fund co-founded by Peter Diamandis), Sinovation Ventures (founded by Dr. Kai-Fu Lee), A-Level Capital, billionaire Jim Mellon, Qiming Venture Partners, Pavilion Capital, Baidu Ventures, F-Prime Capital, Lilly Asia Ventures, Eight Roads, WuXi AppTec, and Juvenescence. Insilico Medicine has garnered many industry awards and recognition including the NVIDIA Top 5 AI Companies for Social Impact in 2017, the Frost & Sullivan North American Technology Innovation Award in 2018, and CB Insights’ AI 100 annual ranking of the 100 most promising AI startups in the world in 2018.
The interview has been edited and condensed.
Cami Rosso: On May 11, 2020, Insilico Medicine announced the results on generating novel potential chemotypes targeting SARS-CoV-2 main protease (Mpro). Can you talk to me about 6LU7 and where you are now that we are in June?
Alex Zhavoronkov: Yes, from 6LU7 we have had some hits. We have had some very promising results.
CR: Why are elderly people more at risk for COVID-19?
AZ: There are multiple hypotheses. It’s always a hypothesis and not a fact until we conduct clinical studies and go very deep into the mechanisms of the disease. My hypothesis is that with age, there is a process called immunosenescence so your immune system function declines.
For example, all of us are born with a thymus where T cells mature. By the age of 30, the thymus usually stops working properly and by the age of 40 you typically lose the thymus—it’s just a bunch of fibrotic tissue sitting there.
The immune system in general loses its potency with age because evolution really needs us there to reproduce and take care of our young, then gracefully decline. We’re not built to last forever and to last substantially past that reproductive period. So technically, us living to 70 is very unnatural. The immune system just declines over time.
CR: Tell me about Insilico Medicine’s Lego-like system.
AZ: The system is structured as three major building blocks: identification of molecular targets, generation of novel small molecules, and prediction of clinical trials. For generative chemistry, it’s also many building blocks, some involve virtual screening.
Many of our “competitors” just go through multiple molecular libraries for prioritizing molecules based on their predicted capabilities. They are taking existing molecules from a library, not making them from scratch. That’s significantly cheaper than synthesizing, so we have those blocks as well. Some of those blocks come from transcriptional response.
We also have a structural based scoring engine, which looks at the structure of the actual molecule and makes a prediction, and have pathways-based scoring engines. We published a bunch of algorithms that do it. It also comes from proteomics primarily.
Our crown jewel is a system that generates normal chemical matter using general adversarial networks. That system itself is a combination of many, many building blocks. It’s a huge system with many different models that generate novel chemical matter. It’s based on either structure of the protein (so it looks for pockets and then generate templates in the pockets), or they have ligand-based generation where you show it the molecule that hits the desired target of interest, and it looks at the molecule and generates any variance of this molecule with desired properties.
There is a pipeline where we just take this protein sequence, properly folded, properly structure it, and then use a structure-based approach.
It can handle many cases and uses different approaches to generate different molecules. It can use more than 800 different models that we have. So here you have a Lego system of many, many approaches for generation. But then after you generate, you start prioritizing the generated molecules and you are doing virtual screening on the chemical matter.
CR: So first the system imagines, then prioritizes versus the others who just prioritize from secondary data from a database.
AZ: Exactly. That’s correct. But we can also prioritize from those molecular libraries. We do this when we want to do it cheaply. Chemical synthesis is a very expensive and often a lengthy process. If your GAN generates, let’s say five molecules, and if they are not easy to synthesize, you have to do the work, the several months of the synthesis. But very often if you just arrive at something that is similar in a molecular library, then you order and send it for testing to see if it hits. If it hits very nicely, then you work around that specific scaffold and you generate around that certain scaffold. But if it doesn’t hit, you actually have to synthesize and do the work. But it gives you some hits.
CR: What’s the percentage breakdown of your focus?
AZ: Currently we have nine programs. We are focusing primarily on oncology. About 30-40 percent of what we have are focused on oncology targets. That includes immuno-oncology as well. And then we have 30 percent of what we have are different types of fibrosis. The rest are a spectrum. We have anti-infectives, both antivirals and antibacterial targets. We also identify the normal mechanisms for antibacterial drugs. We have a couple of metabolic targets. And we have CNS. Some of those targets can work for multiple diseases and areas. Some of those targets we identified through our research in aging. Aging is associated with a very large number of diseases. So, we identified some of those targets and we can repurpose into multiple therapeutic areas.
Copyright © 2020 Cami Rosso All rights reserved.
