Researchers at St. Jude Children’s Research Hospital, Harvard Medical School, and additional institutions have unveiled AdaptiveFlow, an AI-powered platform designed to evaluate billions of drug-like compounds at a cost far lower than existing techniques. The findings were published in Nature Biotechnology, where the team described the tool as a breakthrough for large-scale virtual drug screening in everyday research settings.
What sets AdaptiveFlow apart is its unprecedented ability to scale across 5.6 million virtual CPUs in cloud environments, a first for drug discovery applications. This linear scaling ensures performance remains stable even when processing enormous molecular libraries—such as the 69 billion compounds tested in the study’s demonstration. Conventional screening methods degrade in efficiency as CPU demands grow, but AdaptiveFlow overcomes this limitation by maintaining consistent speed regardless of scale.
The platform operates through an 18-dimensional grid that organizes molecules by key properties, including molecular weight and structural features. A machine-learning algorithm then filters these candidates, applying over 1,500 docking simulations to assess binding potential. Preliminary tests revealed that the top candidates identified by AdaptiveFlow often outperformed those selected through traditional screening, reducing the time required for later-stage drug refinement.
Read Also: Interferon-Alpha Reduces Harmful Blood Cancer Cells: Study
To validate its effectiveness, the researchers applied AdaptiveFlow to two distinct cancer-related proteins: PARP1, a long-studied target in oncology, and FSP1, an emerging protein associated with programmed cell death. The platform successfully identified inhibitors for both, with the FSP1 results, including binding at a previously overlooked, co-factor-dependent site, demonstrating its flexibility. PARP1, already the focus of approved therapies, served as a benchmark, while FSP1’s complexity showed the tool’s ability to tackle difficult targets.
The developers have made AdaptiveFlow freely available as open-source software, complete with detailed documentation and tutorials. This accessibility removes barriers for smaller research groups and institutions lacking advanced computational infrastructure. By lowering the cost and technical hurdles of ultra-large screens, the platform could improve the efficiency of drug discovery pipelines, particularly for targets where conventional methods have proven inadequate.
The tool is now accessible for download through the project’s official website at adaptive-flow.ai and its GitHub repository at GitHub. The research was directed by Christoph Gorgulla, PhD, a principal investigator at St. Jude’s Center of Excellence for Data-Driven Discovery.
Leave a Reply