A New AI Tool Speeds Up Drug Discovery for Cancer and Beyond

Researchers at Tsinghua University in China have created DrugCLIP, an AI system that finds potential new medicines millions of times faster than old methods. Traditional drug discovery uses slow computer simulations to check if a small molecule fits into a protein’s binding pocket, the spot where drugs attach to block or change a protein’s action in disease. This process can take days or weeks for one protein and millions of compounds, making it costly and limiting for urgent needs like cancer treatments.

DrugCLIP changes this by turning proteins and molecules into math vectors, like points in a digital map. A neural network for proteins and another for molecules learn from known drug-protein pairs, so close points on the map mean good matches without any slow shape-fitting. To handle thousands of proteins, the team used AlphaFold2 (we talked about it here) to predict 3D shapes for about 10,000 human proteins covering half the druggable genome, then refined binding pockets with their GenPack tool for better accuracy. In tests, it screened 500 million compounds against 10,000 proteins (10 trillion pairs) in one day on standard computers.

The tool shone on tough targets like TRIP12, a protein tied to cancer growth and autism, where no good structure existed before. DrugCLIP found strong binders that lab tests confirmed block TRIP12’s activity, a win for “undruggable” proteins. It also nailed known targets like serotonin receptors and transporters, matching or beating docking methods in hit rates up to 17.5% in experiments. Other reports note similar success on diverse proteins, with AI spotting hits overlooked by slower tools.

Freely available to the public, DrugCLIP includes a full database of pre-screened results for researchers worldwide. This open resource speeds hunts for new drugs against rare diseases or hard cancers. As AI builds on AlphaFold, tools like this cut discovery time from years to hours, promising faster therapies without hype, just real lab-backed speed.

Source.