Effective AI in drug discovery hinges on trust

In late September, I attended “Data & AI to Accelerate Drug Discovery,” an event in London organized by CAS (the information arm of the American Chemical Society, which also publishes C&EN) and AstraZeneca. Many of the key themes discussed feature in my colleague Aayushi Pratap’s latest story, which is the cover story of this month’s print issue. She reports on how pharmaceutical companies are increasing their investments in artificial intelligence and machine learning.That is unsurprising. Going from an idea to a new drug that is treating patients is a long, expensive business. Indeed, most of the time, these campaigns fail for a myriad of reasons. Which is why the promise of speeding up various areas of drug discovery is so appealing to business. From building the code that will run simulations to suggesting new drug targets or molecules for experimental assays, machine learning and large language models are quietly finding real use in the pharmaceutical industry.But nobody at the event I attended last month was talking about a human-free future in which a computer runs a whole drug discovery program for them. There are a couple of reasons for this. Culture is keyMany people in the room shared that much of their internal data are in silos and not ready to be used by these models. In fact, much of the discussion centered on how to build and manage data sources that could be useful for training the next generation of programs and algorithms. For instance, we need cross-industry benchmarking