Open Discovery Challenge: Four Diseases, Four Targets, One Verifier
aidrugdiscoverysciencebenchmarks
TL;DR
The Open Discovery Challenge asks for one thing: a molecule, written as SMILES. Every entry goes through the same automated verifier and lands on a public leaderboard.
Seasons
| Season | Disease | Target | Counter target | Closes | Prize |
|---|---|---|---|---|---|
| 1 | Malaria | PfDHODH | human DHODH | 2026-09-30 | USD 1,000 (closed) |
| 2 | Tuberculosis | InhA | human FASN (ER domain) | 2026-10-31 | USD 2,000 |
| 3 | Chagas disease | T. cruzi CYP51 | human CYP51A1 | 2026-11-30 | USD 1,000 |
| 4 | Non-opioid analgesia | Nav1.7 (VSD4) | hERG | 2027-01-31 | USD 1,000 |
The counter target is the point. A molecule that binds the human protein as strongly as the pathogen's does not win on potency alone.
How the verifier scores
Six axes: activity, binding, selectivity, ADMET, novelty and synthesis. The weights change by season because the biology changes. For the analgesia season, binding and selectivity carry 30 points each, because a sodium-channel drug that also hits hERG is a cardiac risk.
Every entry is docked repeatedly and the spread is reported with the score, so one lucky pose cannot decide a rank. If any axis cannot be computed, no score is written: a missing axis is a state, not a zero.
What stays private
The scoring code, the gate thresholds beyond the published entry limits and the novelty reference sets. Publishing them would let an entry be shaped against the grader rather than the biology.
Facts and entry limits: github.com/final-bench/open-discovery-challenge
These are computational predictions, not experimental measurements. A high score is a hypothesis worth testing in a laboratory.