AI-Enforced Ultra-Large Virtual Screening Discovers Potent CD28 Binders

Area of Science:

• Computational chemistry
• Immunology
• Drug discovery

Background:

• Targeting protein-protein interactions (PPIs) with small molecules is difficult due to their challenging binding sites.
• Traditional structure-based virtual screening (SBVS) faces computational limitations with large chemical libraries.

Purpose of the Study:

• To demonstrate the first real-world application of the AI-powered PyRMD2Dock platform for structure-based virtual screening.
• To discover novel small-molecule modulators of the immune receptor CD28.

Main Methods:

• Utilized PyRMD2Dock, an AI-enhanced SBVS workflow, to screen 2.4 million compounds against CD28.
• Trained machine learning models on initial docking data to rapidly screen an additional ~46 million compounds.
• Prioritized ligands based on interaction filtering and clustering.

Main Results:

• Identified multiple direct CD28 binders from the screened chemical library.
• Developed lead compounds (100 and 104) with submicromolar affinity (K_d = 343.8 nM and 407.1 nM).
• Demonstrated potent disruption of CD28-CD80 interaction and functional blockade in cellular assays, reducing cytokine secretion.

Conclusions:

• PyRMD2Dock is a highly scalable and effective platform for identifying small-molecule modulators of challenging targets like immune receptors.
• This AI-driven approach overcomes computational bottlenecks in mining massive chemical libraries for drug discovery.