(Yicai) Jan. 13 -- A team of scientists at Tsinghua University in China have developed an artificial intelligence-driven virtual drug screening platform capable of operating up to 10 million times faster than conventional molecular docking methods, enabling genome-wide drug discovery at unprecedented speed.

The team, led by Lan Yanyan from the university's Institute for AI Industry Research, has developed DrugCLIP, a contrastive learning-based framework for rapid and accurate virtual screening. It enables trillion-scale screening across the human druggable proteome and provides an open-access resource that lays the groundwork for next-generation drug discovery, according to a study published in the academic journal Science on Jan. 8.

At present, targeted drugs exploration only covers about 10 percent of known druggable targets in the human body, so researchers have begun using AI tools to screen candidates and accelerate new drug development. For example, Google DeepMind’s AlphaFold has transformed protein structure prediction. But such tools are still too computationally intensive to cover genome-wide targets.

DrugCLIP addresses this computational bottleneck by encoding protein binding pockets and small-molecule compounds into a shared latent space, the study said. The model is trained using a combination of large-scale synthetic data and experimentally determined protein-ligand complex structures, an approach that allows large compound libraries to be rapidly searched against protein targets using dense retrieval techniques similar to those employed by modern search engines.

To demonstrate DrugCLIP’s efficiency, the Lan’s team conducted a genome-wide virtual screening involving about 10,000 human proteins and 500 million compounds. The system evaluated more than 10 trillion protein-ligand pairs in under 24 hours using only eight graphics processing units. The screening identified more than 2 million candidate molecules across nearly 20,000 binding pockets, covering roughly half of the human genome.

Building on these results, the team released GenomeScreenDB, the world’s largest protein-ligand screening database. The database is freely accessible to the global research community and is intended to provide robust data support for basic research and early-stage drug discovery.

Editor: Martin Kadiev