Researchers from Peking University (China), the University of Washington (US), and the AI tech firm INF Technology (China) have introduced ActFound, a novel AI-powered drug discovery model.

The ActFound model has been trained using a collection of assays from a widespread chemical database and an array of measured bioactivities. It employs two ML techniques: Meta-learning and pairwise learning. Researchers claim that ActFound presents a less costly and precise alternative to more traditional methods through its capacity to perform accurately with fewer data points and could serve as a reliable model for predicting a range of bioactivities. 

Analyst QuickTake: In an article published in Nature , the research team demonstrated the ActFound model's functionality. It is a bioactivity foundation model designed to learn the relative bioactivity differences between two compounds within the same assay and identify incompatibilities among assays. Compared to other AI drug discovery models, which focus broadly on drug discovery processes, such as identifying novel compounds, predicting drug-target interactions, or optimizing pharmacokinetics, ActFound takes a specialized approach to undruggable targets often overlooked by conventional models. Researchers also claim that ActFound is an accurate alternative to the physics-based computational tool FEP+(OPLS4) performance-wise, using only a few data points for fine-tuning.