The Hidden Cost of "Innovative": Rethinking How We Regulate Computational Drug Discovery
Touring Schrödinger's drug discovery platforms was the first time I saw the front end of the pipeline I study from the back end. In my own research, I look at the biology after the fact — how immune cells and microbial signals behave inside an existing tumor. Watching a computational platform screen enormous libraries of molecular compounds to identify potential drug targets was a look at the opposite end: software finding candidates faster than a lab ever could on its own.
That speed is real, and it's not hype. Algorithms can analyze massive amounts of biological data at once, uncovering relationships between molecules and targets, and screening candidate compounds at a scale no team of chemists could match manually. That's the "innovative" part everyone talks about.
What gets talked about less is that regulation is only just catching up to it. The FDA issued its first draft guidance on AI in regulatory decision-making for drugs in January 2025. A year later, in January 2026, the FDA and EMA jointly published ten guiding principles for AI in drug development — the first time the two agencies have aligned on AI in pharma. One of the clearest requirements to come out of that: AI tools have to be explainable and interpretable. Drug developers can no longer submit an application arguing, in effect, "the computer thinks this molecule will work" — they have to be able to show why. As of now, no fully AI-developed drug has been approved by the FDA, though the first one is projected within the next year or two.
That timeline matters for a simple reason: the tools are already being used to find and design drug candidates today, years before the regulatory framework governing them was finalized. Speed at the discovery stage is a genuine scientific gain. But "faster to find a candidate" says nothing about "faster to make it affordable" — and if regulation is still being written for a technology that's already screening thousands of compounds a day, there's a real risk that the rules get finalized around whatever the industry has already built, rather than the industry building around rules that account for access from the start.
BioPolicy's interest in this isn't skepticism about the science — it's a question about sequencing. If computational platforms are going to reshape how fast new cancer treatments get discovered, the policy conversation about who can afford what comes out the other end needs to happen at the same speed, not five years after the fact.