AI-driven platform enables billion-molecule virtual screens for rapid drug discovery
Nature Biotechnology introduces AdaptiveFlow, an open-source platform that enables AI-guided ultralarge virtual screenings of 69 billion Enamine REAL Space molecules. It uses an 18-dimensional property grid and Adaptive Target-Guided Virtual Screens to efficiently narrow chemical space, with optional active learning to further accelerate prioritization. The suite supports ~1,500 docking protocols, GPU/CPU cloud scaling (demonstrated up to 5.6 million CPUs on AWS), SELFIES and stereoisomer enumeration, and ML-assisted triage to reduce docking calls. Experimental validation yielded nanomolar inhibitors for FSP1 and PARP1 with structural insights, showcasing scalable AI-driven drug discovery.