Physics-derived geometric scoring. No neural networks. Validated benchmark runs play deterministically in your browser.
Docking
Affinity
Select a benchmark to play through the validated production result.
Validated run: Results from the production artifact.
Create an account to run live scoring on your own structures.
What is this page?
Select a benchmark above to play through the validated production artifact in your browser. The page fills the same progress, feed, verdict, and target table from frozen benchmark outputs, so public visitors can verify the published numbers without loading private scorer bundles.
What do these benchmarks test?
These are the standard retrospective virtual screening benchmarks, chosen to span straightforward discrimination through fully experimental data.
DUD-E, 87 targets. Active versus decoy discrimination. Each target pairs known binders with property matched decoy molecules, and the score is how well actives rank above decoys (AUC).
DEKOIS2, 76 targets. The same discrimination test with harder decoys. Decoys are more tightly physicochemically matched to the actives, which removes easy shortcuts and makes enrichment tougher to achieve.
LIT-PCBA, 15 targets. Built from real high throughput screening confirmatory assays, so actives and inactives are experimentally measured rather than computationally generated. This is the most realistic and the hardest of the three.
Running benchmark...
Target: 0/0
Elapsed: 0s
Lig/sec: --
Completed: 0
Running mean IsoScreen AUC:--
Benchmark progress will appear here as targets complete...
Completed: 0/0Mean AUC: --Lig/sec: --
Deterministic validated-run playback. Results match the production benchmark artifacts.
Your Result
--
vs
Published
--
Per-target results
Per-target Results
Target ↕
AUC ↕
Time ↕
How were these computed?
Scorer version--
Hardware--
Determinism--
--
--
Want to score your own compounds?
Create a free account to run live docking in your browser, same engine, your structures, real-time results.