DrugCLIP opens a new chapter in drug discovery with genome-scale virtual screening.
Beijing, China, 9 January 2026 – A new artificial intelligence system is changing how scientists discover medicines, making it faster, broader, and more accessible than ever before. Known as DrugCLIP, this next-generation virtual screening engine can identify potential drug targets across the entire human genome at a scale and speed that was once considered impossible.
The technology, recently outlined in Science, marks the first successful genome-wide virtual screening for human drug targets. DrugCLIP screened more than 10,000 human proteins against a massive library of 500 million chemical compounds, creating a powerful new resource for drug research worldwide.
Developed by a research team led by Lei Liu, PhD, at Tsinghua University, DrugCLIP uses an advanced AI contrastive deep learning framework to rapidly match small molecules with protein targets. Instead of testing compounds one by one, the system analyzes them all at once using a shared mathematical space, dramatically reducing time and computing effort.
“From target to clinic, DrugCLIP is shortening the distance to hope,” the researchers said. “This is where AI meets drug discovery and becomes the starting point of next-generation drug development.”
Breaking the screening bottleneck
Drug discovery has long faced a major challenge. While AI has improved how drugs are designed, traditional virtual screening methods remain slow. Screening one billion compounds for a single protein target using standard docking techniques can take more than two weeks, even with access to thousands of high-powered computer processors. Doing this for thousands of proteins becomes impractical.
DrugCLIP removes this bottleneck. It uses contrastive learning to encode protein pockets and small molecules into a shared digital space. This allows the system to instantly retrieve promising drug candidates without performing time-consuming one-to-one docking simulations.
Using only eight graphics processing units, DrugCLIP evaluated more than 10 trillion protein–compound pairs in under 24 hours. This achievement places virtual screening firmly into the ultra-high-throughput era.
A genome-wide view of drug possibilities
The system was applied to nearly 10,000 human proteins and identified over two million candidate molecules across about 20,000 binding pockets. This represents roughly half of the known human genome and provides one of the most detailed maps yet of potential druggable targets.
DrugCLIP also performed well in challenging conditions. It outperformed traditional docking and machine learning approaches even when working with incomplete protein structures, noisy data, novel targets, Apo proteins, and structures predicted by AlphaFold. This makes it especially valuable for early-stage research where experimental data is often limited.
Real-world validation
To test whether the system’s predictions work beyond computers, the researchers conducted biological validation using the norepinephrine transporter, an important target in mental health treatment. Around 15 percent of DrugCLIP’s predicted compounds were confirmed as effective inhibitors in laboratory testing. Notably, 12 compounds showed stronger binding activity than bupropion, a commonly prescribed antidepressant.
These results suggest that DrugCLIP is not only fast but also accurate, helping researchers identify high-quality drug candidates with a higher chance of success.
Open access for global research
In a move aimed at accelerating innovation, the team has made the entire genome-scale screening database freely available at drugclip.com. Researchers can explore results with one-click access and no coding skills required. This openness allows scientists around the world to build on the findings, speeding up discovery across many disease areas.
The future of AI-driven drug discovery
The researchers believe DrugCLIP represents more than a single tool. It points toward a future where AI-powered screening, structural modeling, and affinity prediction work together to systematically explore the human genome for new therapies.
“The integration of ultrafast virtual screening with emerging prediction technologies will provide a foundation for accelerating future drug discovery efforts,” the team concluded.
