AI biotech: separating signal from synthetic noise.
Most AI biotech pitches collapse under one question. The companies that survive share four structural traits we underwrite for.
The single question we ask every AI biotech founder: what proprietary data asset, not available to a well-funded incumbent, makes your model materially better at the specific decision you are trying to automate?
Four structural traits
1. A defensible data flywheel — typically wet-lab generated, not scraped. 2. A wet-dry loop where model output is empirically validated and the result is fed back into training. 3. A go-to-market that does not require persuading conservative pharma R&D heads to change their internal workflow on faith. 4. Founders who are scientists first and AI practitioners second, not the inverse.
Where the alpha sits
We see the highest expected return in narrowly scoped models tied to specific decision points — target selection inside one modality, PK/PD prediction for one chemical series, patient stratification for one indication — rather than horizontal 'foundation model for biology' bets.