In the vast chemical space, efficiently and precisely identifying molecules capable of treating specific diseases stands as a core challenge in novel drug discovery. To overcome this hurdle, DNA-encoded library (DEL) technology emerged.

DEL technology tags trillions of compounds with unique identifiers. This enables researchers to precisely identify molecules from this massive library, rapidly obtaining hit compounds. With immense potential, DEL has become one of the most highly anticipated early-stage drug discovery technologies in recent years.

However, despite over two decades of development, the number of molecules discovered using DEL technology that have entered clinical trials remains extremely limited. Practical applications have shown significant variations in the hit rates of DEL technology for different target proteins, which has become a critical bottleneck hindering its full potential. The underlying reasons for this phenomenon have long remained a knowledge gap within the industry.

Recently, Westlake Pharma co-founders Dr. Huang Jing and Dr. Hu Qi published a research breakthrough in Nature Communications, titled " Deciphering DEL pocket patterns through contrastive learning."

This study is the first to propose and answer the scientific question of "which protein binding pockets are more suitable for DEL screening" at the binding pocket level. Concurrently, they developed model ErePOC（Enhanced representation of POCkets）, a target protein pocket representation model. It is the industry's first AI model specifically designed to identify DEL-suitable targets.

The study systematically elucidated the key reasons behind the varying efficiencies of DEL screening, laying a crucial theoretical foundation for the industry to apply and optimize DEL technology more effectively in the future.

Through a systematic analysis of large-scale protein pocket structures and DEL binding data, Dr. Huang Jing's team discovered that while DEL-binding pockets significantly overlap with traditionally druggable pockets in overall characteristics, there are previously unrecognized, distinct differences in pocket volume, hydrophobicity, and interaction patterns. Therefore, DEL target selection is not merely a simple extension of the traditional druggability concept but is governed by more specific structural and functional constraints at the pocket level. This study theoretically defined DEL screening target suitability for the first time.

Addressing the challenges of high-dimensional pocket features and sparse DEL data, Dr. Huang Jing's team introduced a contrastive learning strategy. They use ligand chemical similarity as a supervisory signal to guide the compression and alignment of pocket representations, thereby developing the protein pocket representation model ErePOC.

Model ErePOC 's uniqueness lies in its ability to align ligand chemical space with protein pocket functional space for the first time, achieving structure-agnostic target identification. It can assist researchers in discovering potential bindable targets across protein families and is scalable for application in proteome-wide screening.

In this study, Dr. Huang Jing's team validated model ErePOC 's discriminative power and generalization performance by completing multiple downstream tasks.

With model ErePOC，researchers can select targets more suitable for DEL molecules even before initiating DEL screening, thereby focusing resources on directions with higher success probabilities and transforming DEL technology from high-throughput trial-and-error into strategic screening.

The deep integration of DEL and AI represents a frontier in global novel drug discovery and is one of the technological pathways embraced by Westlake Pharma. Early in the development of the company's first pipeline, the team successfully addressed the issue of DEL screening results being cluttered with false positives by leveraging transfer learning and pre-trained models, completing hit compound discovery in approximately two months.

Since then, Westlake Pharma has continuously upgraded its "DEL+AI" R&D model, enabling the team to achieve ongoing efficiency breakthroughs.

As a technology commercialization enterprise stemming from Westlake University and Westlake Laboratory, Westlake Pharma has always rooted its R&D system in a profound understanding of scientific mechanisms. The company's R&D team deeply understands that true industrial breakthroughs cannot merely reside at the application level of cutting-edge technologies but require exploring the fundamental essence behind observed phenomena. This theoretical research achievement serves as a solid foundation for Westlake Pharma 's "DEL+AI" technology iteration.

Efficiently translating advanced technologies into novel drugs with clinical value is Westlake Pharma 's unwavering mission. The realization of this ambitious goal also relies on industry-wide progress. Westlake Pharma anticipates that the founder team's current research findings will bring new insights to the industry, and looks forward to collaborating with industry partners to achieve breakthroughs in efficiency, ultimately allowing innovative medicines to reach patients faster.