question · intermediate · partly written
Could one model cover every step of drug discovery?
Whether one model could cover every step of drug discovery; the sourced material here bears on baselines and on how little a single readout predicts.
What the models-of-models material says more
Cell-based data types each predict 6 to 10% of 270 assays with high accuracy and 21% combined. In perturbation prediction, a simple linear baseline did not lose to deep-learning foundation models on double perturbations in one study, and the Virtual Cell Challenge organisers say models do not yet consistently outperform naive baselines across all metrics. When a model predicts a model, the baseline should be checked first, because the readout beneath both is itself a model. 123
Reported, not yet replicated
Still open here
- What does machine learning do at each step today, how analogous are the steps, and what stands between that and a single model?
Not yet written: Observation, Hypotheses, What is missing, What is known, What is not, What would settle it, Who works on it.
Linked from
Sources
- Moshkov et al.. Predicting compound activity from phenotypic profiles and chemical structures (Moshkov et al. 2023), Europe PMC listing. Europe PMC listing (2023). paper; opened 2026-11-17 all records citing this
- Ahlmann-Eltze C, Huber W, Anders S. Deep-learning-based gene perturbation effect prediction does not yet outperform simple linear baselines. Nat Methods (2025). paper; opened 2026-11-17 all records citing this
- Virtual Cell Challenge 2025 wrap-up. Arc Institute (2025). news, not peer reviewed; opened 2026-11-17 all records citing this
- Amit's dictation of 2026-11-17: what the atlas must contain. Kept in the atlas repository; not published.
History of this record (2)
This record has been changed 2 times, on the dates below. The note kept against each change is written for the people who maintain the atlas, in their own shorthand, so it is not shown here; it is in the record itself.
- Kind
- amit
- Register
- open-question
- Status
- partly written
- Last updated
- 2026-11-17
- Record
- content/questions/one-model-to-rule-them-all.json