Better medicines thanks to AI? Insitro CEO explains what machine learning can teach big pharma

WASHINGTON– Artificial intelligence is changing the way companies do business: it helps programmers write code and answer customer service calls with chatbots.

But the pharmaceutical industry is still waiting to see if AI can solve its biggest challenge: finding faster, cheaper ways to develop new drugs.

Despite the billions invested in research, developing new drugs still typically takes a decade or more.

Founded in 2018, Insitro is part of a growing group of AI companies promising to accelerate drug discovery by using machine learning to analyze huge data sets of chemical and biological markers. The South San Francisco-based company has signed deals with drugmakers like Eli Lilly and Bristol Myers Squibb to help develop drugs for metabolic diseases, neurological conditions and degenerative disorders.

CEO and founder Daphne Koller spoke with the AP about what AI brings to the challenges of drug discovery. The conversation has been edited for length and clarity.

A: I think the problem with drug discovery is that we’re trying to intervene in a system that we barely understand. Many of the successes we’ve had over the last 15 to 20 years have come from understanding the system well enough to actually design interventions that fit within it.

So one of the things we’re trying to do at Insitro is unravel the underlying complexity of heterogeneous diseases and identify new modes of intervention that might help, perhaps not the entire population, but perhaps only a subset of it. In this way, we can actually identify the right therapeutic hypothesis to intervene in a particular patient population. And that, I think, is the real crux of the industry’s lack of success.

A: One of the things that’s happened alongside the AI ​​revolution is a much quieter revolution in what I call quantitative biology, which is the ability to measure biological systems with unprecedented fidelity. You can measure systems like proteins and cells with increasingly powerful measurements and technology.

But if you give this data to a person, their eyes will simply be dazzled, because there are only so many cells a person can look at and only so many subtleties they can see in these images. . People are simply limited in their ability to perceive subtle differences.

So you end up with a very reductionist view of a very complex and multifaceted system that is really important for untangling distinctions between patients and discovering where an intervention can actually make a difference.

A: My doctorate was in computer science. But I became interested in the field of machine learning for biomedical problems in 1998 or 1999.

At that time, the problems that machine learning was capable of solving were frankly uninspiring. How much inspiration can you use to classify spam versus non-spam in a dataset of email messages?

I was looking for something richer. And my first foray into this field was not because I was particularly interested in becoming a biologist, but because I was looking for more technically challenging questions. And then, as I got deeper into it, I became interested in biology in its own right.

A: It’s probably one of the most important things we’ve accomplished as an organization.

You can take the most sophisticated and sensible scientists from both sides and put them in the same room together and they might as well speak Thai and Swahili to each other.

When you are an engineer, you look for the strongest and most consistent models that will allow you to make predictions about a majority of cells or individuals. When you’re a life scientist, you often look for exceptions because those are the things that can lead to new discoveries.

So we have put in place a number of cultural and organizational elements to help people interact with each other in an open, constructive and respectful way.