Graph_RG: Dominating CASP16's Small Molecule Affinity Prediction Subcategory-A Pose-Free Framework for Billion-Scale...

Area of Science:

• Computational chemistry and cheminformatics
• Artificial intelligence in drug discovery
• Biomedical data science

Background:

• Protein-ligand interaction prediction is crucial for early drug development stages, including virtual screening and target identification.
• Existing methods often rely on complex conformational searches, limiting their scalability and efficiency.
• Developing accurate and computationally efficient models is essential for accelerating drug discovery pipelines.

Purpose of the Study:

• To present Graph_RG, a high-performing model for protein-ligand affinity prediction.
• To demonstrate the computational efficiency and scalability of Graph_RG compared to existing methods.
• To explore potential improvements and broader applications of Graph_RG in drug discovery.

Main Methods:

• Development and application of the Graph_RG model for protein-ligand affinity prediction.
• Performance evaluation in the CASP16 small molecule track.
• Benchmarking computational efficiency against conformation-search dependent methods.

Main Results:

• Graph_RG achieved the top performance in the CASP16 protein-ligand affinity prediction category with an N-weighted Kendall's Tau of 0.42.
• The model demonstrated exceptional computational efficiency, operating over 100,000 times faster than traditional methods.
• Enabled large-scale virtual screening (billions to 10 billion compounds) on standard servers.

Conclusions:

• Graph_RG offers a highly accurate and computationally efficient solution for protein-ligand interaction prediction.
• The model's speed and scalability facilitate large-scale drug screening and target identification.
• Future work includes model optimization and development of an accessible web platform to integrate AI into practical drug discovery.