(Yicai) Jan. 13 -- A  scientific research team from China's Tsinghua University has developed a genome-wide virtual screening platform that uses artificial intelligence for drug discovery, which is up to 10 million times faster than traditional methods such as molecular docking.

The team, led by Lan Yanyan from Tsinghua University's Institute for AI Industry Research, developed the contrastive learning framework DrugCLIP for fast and accurate virtual screening, enabling trillion-scale screening covering the human druggable proteome and providing an open-access resource that forms a foundation for next-generation drug discovery, according to a study published in Science on Jan. 8.

The exploration of targeted drugs only covers about 10 percent of all druggable targets in the human body, so researchers began using AI tools to screen candidates and accelerate new drug development. For example, Google's DeepMind developed the protein structure-predicting model AlphaFold, but it and other such tools are still too computationally expensive to cover genome-wide targets.

To address this bottleneck, DrugCLIP encodes protein pockets and small molecules into a shared latent space, trained using both large-scale synthetic data and experimentally determined protein-ligand complex structures, the study said. Large compound libraries can then be rapidly queried with protein targets with dense retrieval techniques, similar to modern search engines.

The team applied DrugCLIP to a genome-wide virtual screening on about 10,000 human proteins against 500 million compounds to validate its efficiency, scoring more than 10 trillion protein-ligand pairs in under 24 hours using only eight graphics processing units. The screening yielded more than two million candidate molecules covering nearly 20,000 pockets, representing around half of the human genome.

Based on this result, the team released the largest protein-ligand screening database in the world, GenomeScreenDB, making it free to access for the global research community to provide strong data support for basic research and early drug discovery.

Editor: Martin Kadiev