AI Structural Prediction Accelerates Molecular Glue Drug Discovery

Researchers at Baylor College of Medicine have accelerated the discovery of molecular glue degraders through an integrated approach combining high-throughput proteomics and artificial intelligence. The findings, published in Nature Communications in 2026, detail a novel class of small molecules capable of targeting and eliminating VAV1, a critical regulator of immune cell signaling implicated in T-cell lymphomas and chronic inflammatory disorders. Led by Dr. Jin Wang and Dr. Hanfeng Lin, the team initiated the project by screening extensive molecule libraries using data-independent acquisition proteomics. This unbiased analysis identified compounds that selectively reduced VAV1 levels, with follow-up assays confirming degradation via the cellular machinery protein cereblon. A primary obstacle in molecular glue development has been predicting the precise structural interactions between the glue, its target, and cereblon. To overcome this, the researchers developed GluePlex, a computational workflow that merges artificial intelligence-based structure prediction with physics-based modeling. Operating without prior experimental structural data, GluePlex accurately predicted the formation of a three-component complex and isolated a specific surface loop within the VAV1 SH3-2 domain as the essential binding and degradation signal. This discovery challenges the prevailing assumption that cereblon-dependent degraders strictly require a G-loop, effectively expanding the targetable proteome. Following the computational breakthrough, medicinal chemists refined the lead compounds to enhance stability and efficacy. By introducing chlorine atoms to reduce molecular flexibility, the team generated NGT-201-18, a highly potent degrader that significantly strengthened the target complex. Subsequent testing in primary human T cells demonstrated that NGT-201-18 successfully lowered VAV1 concentrations and suppressed downstream immune signaling, validating the compound's biological relevance and therapeutic potential for autoimmune conditions. The study underscores the transformative capacity of artificial intelligence to compress traditional drug discovery timelines. By deploying predictive structural modeling at the earliest development stage, the Baylor team bypassed months of iterative experimental screening. The researchers also highlighted the necessity of comprehensive proteomic profiling, noting the unintended degradation of LIMD1 by certain compounds, a reminder that off-target assessment remains critical. Collectively, this work establishes a reproducible pipeline for accelerating the development of targeted protein degradation therapies, offering a promising avenue for treating immune-mediated diseases and hematological malignancies that currently lack effective interventions.