An accurate nucleic acid-small molecule docking framework via geometric deep learning with large-scale pretraining
Researchers have developed NucleoDock, a new deep learning framework designed to improve the accuracy and efficiency of docking small molecules to nucleic acid structures. This method addresses the challenge of limited experimental data by combining large-scale pretraining on synthetic complexes with fine-tuning on real-world data. NucleoDock integrates sequence and structural information with 3D features, utilizing a geometric scoring head for pose ranking. In benchmarks, NucleoDock demonstrated superior performance compared to traditional methods, achieving a 56% success rate in identifying correct docking poses and significantly improving early enrichment in virtual screening. AI
IMPACT Enhances computational drug discovery by improving the accuracy and speed of identifying potential therapeutic molecules for nucleic acid targets.