Uni-Dock

Uni-Dock performs GPU-accelerated molecular docking for structure-based virtual screening of ultralarge compound libraries.

Key Features:

• GPU Acceleration: Uses GPUs to accelerate docking, reporting greater than 1000-fold speedup compared to a single-core CPU run of AutoDock Vina.
• Scoring Functions: Supports vina, vinardo, and ad4 scoring functions and implements hydrogen bond biased docking across all supported scoring functions.
• Batch Processing with Concurrent Threads: Processes molecules in batches with concurrent per-molecule threads to optimize GPU–CPU data flow and minimize CPU bottlenecks.
• Multi-GPU Support: Compatible with multiple GPUs across different architectures and manufacturers to enable scalable execution.
• Hyperparameter Optimization: Performance was analyzed on CASF-2016 and DUD-E, yielding three recommended hyperparameter combinations for different docking scenarios.

Scientific Applications:

• Hierarchical virtual screening: Enables hierarchical virtual screening experiments for drug discovery.
• Ultralarge-library screening: Demonstrated screening of the Enamine Diverse REAL druglike set (38.2 million molecules) against KRAS G12D in 12 hours using 100 NVIDIA V100 GPUs.

Methodology:

GPU-accelerated docking with batch processing and concurrent per-molecule threads; support for vina, vinardo, and ad4 scoring functions with hydrogen-bond-biased docking; multi-GPU execution; and hyperparameter analyses on CASF-2016 and DUD-E resulting in three recommended parameter sets.