Transparent acceleration of large library docking with ChemSTEP
Тип публікації :
Препринт
Дата випуску :
18 березня 2026 р.
Автор(и) :
Olivier Mailhot
Katie Holland
Lu Paris
Dmytro S. Radchenko
Yurii S Moroz
John J. Irwin
Brian K. Shoichet
eKNUTSHIR URL :
Журнал :
ChemRxiv
Цитування :
[APA 7] Olivier, M., Katie, H., Lu, P., Dmytro, S. R., Yurii, S. M., John, J. I., & Brian, K. S. (2026). Transparent acceleration of large library docking with ChemSTEP. ChemRxiv,. https://doi.org/10.26434/chemrxiv.15001061/v1
[ДСТУ] Transparent acceleration of large library docking with ChemSTEP / M. Olivier та ін. ChemRxiv. 2026. DOI: 10.26434/chemrxiv.15001061/v1 (дата звернення: 11.09.2026).
While make-on-demand libraries now span trillions of molecules, full library docking struggles beyond a few billion, motivating prioritization strategies that recover top-scoring compounds while evaluating only a fraction of a library. Here we introduce a transparent prioritization approach, ChemSTEP (Chemical Space Traversal and Exploration Procedure), and define the effective size of a library treated by any prioritization algorithm, N eff . Grounded in ligand similarity, ChemSTEP docks a representative seed set, selects diverse high-scoring "beacons", and iteratively traverses the full library through cycles of beacon selection and similarity search. In retrospective calculations on eight targets and six billion molecules, ChemSTEP recovers over 75% of high-scoring compounds while docking less than 5% of the library. We then test ChemSTEP prospectively against AmpC β-lactamase using a 13.2 billion molecule library. Because AmpC recognizes anionic inhibitors, we explicitly docked all 360 million library anions, synthesizing and testing 241 high-ranking ones in parallel to the ChemSTEP 13.2B run; this serves as ground truth for the entire 13.2B enabling replicate ChemSTEP runs and direct comparison to brute-force campaigns. Compared with previous docking of 99 million and 1.7 billion molecules against AmpC, the 13.2 billion library had higher hit-rates (12% vs 41% vs 60%, respectively) and found 80% of the 241 high-ranking compounds within the first 0.5% docked. We estimate that trillion-molecule libraries might be prioritized within a month on a 5k-CPU cluster.
Якщо не вказано інше, ця робота розповсюджується на умовах ліцензії Attribution-NonCommercial 4.0 International