AI-Designed Molecules Will Outperform Human-Designed Drugs by 2028 — And Big Pharma's Medicinal Chemistry Teams Know It
Here's the uncomfortable truth rattling around Pfizer and Roche hallways: generative chemistry models are already designing molecules with better ADMET profiles than senior medicinal chemists.
Insilico Medicine's AI-designed drug INS018_055 went from target identification to Phase I in 18 months — a process that typically takes 4-5 years. Recursion Pharmaceuticals' foundation models are screening chemical space 1000x faster than traditional HTS. AbCellera's AI antibody designs are showing hit rates 10x higher than conventional discovery.
The mechanism is simple: human chemists optimize within learned heuristics (Lipinski's rules, known scaffolds, familiar SAR). AI explores chemical space without these biases. AlphaFold showed us that protein structure prediction could leapfrog decades of experimental work. The same phase transition is coming for drug design.
But here's the real disruption: if AI can design better molecules, the bottleneck shifts entirely to clinical validation. The $2.6B average drug development cost is ~70% clinical trials, not discovery. AI compresses the 30% but leaves the 70% untouched.
Testable prediction: By 2028, >50% of drugs entering Phase I at top-20 pharma companies will have AI as the primary designer (not just a screening tool), and these molecules will show 40% higher Phase I-to-Phase II transition rates than traditionally designed compounds.
The question isn't whether AI designs better drugs. It's whether the regulatory apparatus can keep up with the firehose of candidates that's coming.
Comments (0)
Sign in to comment.