Axios Pro Exclusive Content

Sofinnova's Edward Kliphuis talks AI drug discovery, digital therapeutics

headshot
Nov 27, 2023
a photo illustration of Edward Kliphuis of Sofinnova Partners surrounded by rectangles forming health pluses

Pictured: Edward Kliphuis. illustration: Tiffany Herring/Axios; Photo: Courtesy of Sofinnova Partners

Trial and error isn't just for pharmaceutical companies testing new drugs — it's also an applicable process for the life sciences technology market, says Sofinnova partner Edward Kliphuis.

Driving the news: The venture capital firm last month closed on a $200 million fund to invest across digital medicine and life sciences technology,.

Axios spoke with Kliphuis as part of the Axios Expert Voices series. This interview was edited for length and clarity.

How is Sofinnova viewing the burgeoning AI-powered drug discovery tech market given that some candidates have failed in trials?

  • "The bullish case is we can actually truly do new things that that we could not have comprehended before. We can open up the search space of biology in a way that we haven't done before. Traditionally speaking, sort of you have to compare this to an oil spill, where every new hypothesis really comes to the center out. Every existing hypothesis basically leads to a new one.
  • The bearish scenario, though, is that every single molecule that comes out of these platforms will still have to find its way through the clinic. It still has to be tested in humans at some point.
  • To my knowledge today, there's not a single AI-based or AI-design molecule that actually has been approved.
  • If you look at some of the AI-based drug discovery companies — like a Recursion, an Exscientia — you will actually find that the majority of them have either a zero value to their platform or a negative value to their platform."

The first wave of digital therapeutics saw high-flying unicorns file for bankruptcy and pare down operations. What does the future hold?

  • "We assumed going through a deep and difficult clinical process would ultimately lead to a reward, and that just like with traditional pharmacological drugs, you will be rewarded for the technical risks you take. Turns out that's not the case.
  • We see there is a promise, especially in replacing for example, lower complexity tasks, such as musculoskeletal therapies or the like, where human intervention simply is too costly for the effect size.
  • However, until some of these, these therapies can also be disease-modifying or even curative, it will be very hard to command a high price. So I see two ways out: either is going to be a volume play where you see a digital pharma coming up with a suite of products that ultimately can be commercialized for lower friction, lower hurdle solutions.
  • Or we find digital therapeutics — I'm thinking particularly of the cognitive domain — can have such a sea change effect, maybe on understanding dementia."

What life sciences technology trends do you expect to see in 2024?

  • "I think there'll be a bunch of people trying the large language model approach to biology.
  • You have your base nucleotides, your DNA, which, through translation, transcription can become amino acids. Which become proteins, which have an effector function on different cell types. ... It's a similar sequence of events. So I think we'll see people trying their hand at ... these universal models for biology.
  • We see large language models having hallucinations. What happens if you have a hallucination in a biological large language model? What are the questions you need to ask these models?"
Go deeper