Ligand Pro, a company founded by professors and a PhD student from the Skolkovo Institute of Science and Technology (Skoltech), has developed the MatCha AI model—a fast and accurate method for molecular docking. The model performs virtual screening of potential drug candidates 30 times faster than major co-folding models of the AlphaFold class, while surpassing them in accuracy and physical correctness of results, the Skoltech press service reported.
The technology opens up new possibilities for virtual screening and early-stage drug development, they emphasized.
Disease often arises when one or more proteins in the body stop functioning correctly. In drug development for a given disease, such a “broken” or malfunctioning protein is called a therapeutic target—the drug’s action is directed precisely at it, the scientists explained.
They noted that the goal of a drug is to interact with this target in a way that alters its activity: for example, to “switch off” an overactive protein or, conversely, to “switch on” one that is insufficiently active. Therefore, they said, the correct selection of the drug molecule is crucial.
Marina Pak, co‑founder and CEO of Ligand Pro and a graduate of Skoltech, noted that drug development is a lengthy and capital-intensive process with a high level of risk: a project can be halted at any stage after significant investments of time and resources. At the same time, she said, it is the early stages that are most amenable to optimization using computational methods.
“Over three years, we have progressed from an idea and team formation to a world-class result. We are continuing to develop MatCha, as well as creating tools for related tasks—molecule generation, property prediction and optimisation. Our next step is experimental validation of the technologies in real R&D pipelines, followed by industrial implementation,” said Daria Frolova, Director of Machine Learning at Ligand Pro and a Skoltech PhD student in the Computational Systems and Data Analysis in Science and Engineering programme.
