AdaptiveFlow Opens 69 Billion-Molecule Drug Screens, Finding Nanomolar Inhibitors at 5.6 Million-CPU Scale
Updated
Updated · Nature.com · Sep 1
AdaptiveFlow Opens 69 Billion-Molecule Drug Screens, Finding Nanomolar Inhibitors at 5.6 Million-CPU Scale
2 articles · Updated · Nature.com · Sep 1
Summary
AdaptiveFlow was introduced as an open-source platform for ultralarge virtual screening, packaging a ready-to-dock 69 billion-compound Enamine REAL Space library and workflows for AI- and ML-assisted drug discovery.
An 18-dimensional property grid and optional active learning let the system focus on promising chemical subspaces, cutting docking workloads by orders of magnitude; a 12 million-molecule prescreen can probe the 69 billion-compound library with more than 5,000-fold less computation.
The platform integrates more than 1,500 docking protocols, including GPU-accelerated and ML-based methods, and showed near-linear scaling on up to 5.6 million AWS vCPUs, enabling billion-scale screens in hours rather than weeks.
Experimental validation produced nanomolar inhibitors for two targets: FSP1 hits with Ki values down to 0.283 µM and PARP1 hits including iParp1 with an 8.8 nM IC50, while co-crystal structures confirmed predicted binding modes.
The study positions AdaptiveFlow as infrastructure for screening ever-larger chemical libraries—already expanding toward trillions of molecules—while lowering barriers through open-source code, cloud access and SELFIES-compatible data formats.
Could AdaptiveFlow's 18-dimensional grid and active learning algorithms inadvertently filter out unconventional yet highly effective drug candidates?
Will the ability to efficiently screen billions of compounds on cloud infrastructure finally make traditional brute-force drug discovery obsolete?
How might the adaptive tranche-based search used to explore 68 billion molecules revolutionize massive data sorting in non-medical fields?
Democratizing Ultra-Large Virtual Screening: AdaptiveFlow Delivers 1,000-Fold Cost Reduction for Academic and Small Labs
Overview
AdaptiveFlow, an open-source platform published in Nature Biotechnology, makes ultra-large virtual screening affordable and accessible for academic labs and small startups. By combining a unified automated pipeline with advanced cloud scaling and the Adaptive Target-Guided (ATG) method, it reduces computational costs by up to 1,000-fold and enables efficient screening of billions of compounds. The platform’s design eliminates complex setup, supports major cloud providers, and is engineered for resilience and cost savings. This breakthrough allows researchers worldwide to independently discover and synthesize promising drug candidates, transforming drug discovery and opening new opportunities for neglected and rare disease research.