SparkleDock: GPU-Accelerated Flexible Macromolecular Docking Framework
A new docking framework named SparkleDock has been created by researchers, facilitating near-real-time flexible macromolecular docking on GPU supercomputers. This framework is outlined in a paper available on arXiv (2608.07078) and overcomes the shortcomings of current methods like LightDock, which relies on Glowworm Swarm Optimization (GSO) but faces challenges such as limited parallelism and uneven computation loads. SparkleDock reengineers GSO to reveal extensive fine-grained parallelism at the glowworm-agent level and reformulates energy scoring for Tensor Core compatibility, enhancing the execution of irregular pairwise interactions through structured matrix operations. Additionally, it implements performance-model-driven scheduling for better load balancing and out-of-core scaling across GPUs, achieving speedups of 9.7x and 18.9x compared to existing techniques, marking a significant advancement for computational biology in predicting biomolecular interactions at scale.
Key facts
- SparkleDock is a scalable GSO-based docking framework for GPU supercomputers.
- It enables near-real-time flexible macromolecular docking.
- LightDock, an existing approach, suffers from limited parallelism and load imbalance.
- SparkleDock redesigns GSO for fine-grained parallelism at the glowworm-agent level.
- Energy scoring is reformulated for Tensor Core compatibility.
- Performance-model-driven scheduling handles load balancing and out-of-core scaling.
- Speedups of 9.7x and 18.9x are achieved.
- The paper is available on arXiv with ID 2608.07078.
Entities
Institutions
- arXiv