Turning Biological Complexity into R&D Advantage with Virtual Cells - with Giorgio Gaglia of Sanofi & Daniel Veres of Turbine

Daniel Faggella (0:12)

Welcome everyone to the Emerge AI in Business Podcast. Today's guests are Daniel Veres, Co-Founder and Chief Scientific Officer at Turbine, a company building virtual cell models to stimulate how cells respond to genetic and chemical perturbations, and Giorgio Gaglia, Head of Systems and Disease Biology at Sanofi. Together, they discuss why the most practical unit today is a narrow virtual assay built around a specific experiment, how blind, held-out data builds confidence in a model's predictions, and why the goal isn't to replace the lab, but to quantify the risk behind the next experimental decision. Today's episode is sponsored by Turbine. Just a quick note for our audience that the views expressed by Giorgio in today's program do not reflect that of Sanofi or its leadership. In this episode, we cover what it takes for drug discovery teams to trust a virtual sales output. To go deeper on this topic and learn how consultants are winning business with evidence-based AI ROI and building long-term capabilities instead of chasing short-term gains, download our free PDF report Three Keys to Thriving in the Coming Era of Automation at emerge.com/cok1.

That's emerj.com/cok1 to download your copy. Now the conversation with Daniel and Giorgio.

Giorgio, Daniel, welcome to the Emerge AI in Business Podcast here.

Daniel Veres (1:55)

Welcome.

Giorgio Gaglia (1:55)

Thank you.

Daniel Veres (1:56)

Thank you.

Daniel Faggella (1:56)

So we have some very interesting things to discuss today. But before we get into that, before we kick off things, Giorgio, I know that you have a quick note for our audience that you'd just like to mention.

Giorgio Gaglia (2:05)

Yes. Thank you. I am a Sanofi research employee, but I'm here on my personal stance, and everything I say is mine and mine alone. I also would like to disclose conflict of interest. As a Sanofi employee, I own Sanofi stock.

So thank you for this, and I'm excited to talk to you today.

Daniel Faggella (2:22)

Great. Now that we have transparency out the way we can get into a conversation about trustworthiness. And Daniel, I want to start somewhere that sounds almost too basic, but I really don't think it is. The last few months, I've been seeing the terms of virtual cell and digital twin and simulation all being used almost as if they are interchangeable.

And I don't think they are. And I don't think that the confusion is at a point where it is still harmless. It's really getting to a point where we need to know the differences because this is obviously going to influence how a research leader make their decisions and decide what they're allowed to act on. So before we get into the cleaver stuff, what defines a virtual cell model in practice and why does it deserve to be treated as its own decision layer in drug discovery rather than just the faster lab?

Daniel Veres (3:09)

That's a very good question. And yes, the field and also us sometimes are taking these different definitions. As the virtual cell, I usually say it has a job to do. And that's the main difference compared to like a digital tween or a foundation model.

A virtual cell needs to be useful for some kind of task related to a biological question. So for example, if we are in the drug discovery field, we need to be able to run certain experiments with a virtual cell. And that's actually the definition. There was a very nice paper, the AIVC paper from many leading labs in the field, Endo24, where they defined virtual cells. And there were like three aspects of that definition. But I think the main, the most important, which is discriminating virtual cells from other methods, is that it needs to be able to run in silico experiments.

And that's the job. You are running a certain perturbation. It can be a drug treatment. It can be a genetic perturbation. And you are predicting what's the output of that experiment. So it becomes useful because if this happens at an appropriate level of performance, then you can inform the next decisions, which is usually running the actual wet lab experiment. And that's why we think that it's a decision layer, or it can be, or it is becoming a decision layer of drug discovery. Because today, it's still many times empirical. What are the next experiments? More and more times it's data driven, but we think it should be model driven, and it should be driven by a perturbation engine, which can tell you what's the next experiment. And it should be iterative. So you can set up a real like lab in the loop system, where you are predicting, measuring, and then predicting again if that's needed to reach an appropriate performance. And this in the end leads to smarter decisions.