AI in drug discovery – what it is, where we stand and the path forward
Points and comments are a snapshot, not live.
AI in drug discovery hasn't yet delivered real clinical impact, experts say.
A Nature paper by experts not trying to sell anything concludes evidence for AI's clinical impact in drug discovery is 'disappointingly limited', an absence of evidence, not evidence of absence. Phase II trial success rates are where real improvement would matter, but AI hasn't budged them yet. The authors urge shifting from modeling easy, useless data to generating the hard data needed for real decisions, which is expensive and slow. Key problems: data is too messy for ML, benchmarks don't translate to real-world drug development, and people confuse ligands with drugs.
The field needs more 'why' and less 'because it's new.'
What commenters are saying
Top commenters echo the article's call for moderation: new tech doesn't guarantee outcomes. Some note drug development timelines exceed AI's availability, so measuring impact takes time. Others push back on hype, saying drug scientists already think carefully about their tools. A substantive subthread discusses hair loss treatments: finasteride side effects (2% of users, mostly reversible), topical finasteride, slow-release oral minoxidil (MINX, AI-formulated), and several pipeline drugs (KX-826, clascoterone, VDPHL01, PP-405). One commenter jokes 'AI in drug discovery was never the hard part.' Many praise Derek Lowe's science communication as essential for cutting through hype.