Workshop “Drug design powered by AI: from therapeutic target to small-molecule candidates”

Abstract

The workshop focuses on the use of AI in drug design—from therapeutic target to small-molecule candidates—covering the generation of molecules with desired properties and protein—ligand interaction prediction. It reviews state-of-the-art approaches, including flow matching and other advanced generative architectures, as well as ways to accelerate inference and improve the validity and practical value of predictions, with particular emphasis on filtering and validating AI-generated results in line with medicinal-chemistry considerations and the requirements of real drug-design workflows. Drawing on the AI-driven drug-design platform from the Skoltech-based startup Ligand Pro, the workshop presents the molecular docking neural network Matcha (approximately 31 times faster than large-scale co-folding models from the AlphaFold 3 family), the molecular generator MANGO, and the HEDGEHOG and Bento benchmarks for evaluating molecular generators and molecular docking, respectively; particular attention is given to benchmarking, the integration of multiple tools within a unified environment, and the application of ML to practical drug-development tasks.