Background
Molecular docking is one of the core — and most expensive — steps in drug design: essentially an optimisation problem driven by a scoring function. AutoDock Vina is the most popular member of the AutoDock family thanks to its accuracy and speed, yet unlike its predecessor AutoDock4, hardware acceleration of Vina had rarely been reported.
Approach
Vina-FPGA exploits low-level parallelism in hardware:
- Fixed-point quantization of the docking algorithm, analysed and implemented for better energy efficiency in hardware.
- Multiple in-module hardware pipelines to boost the performance of module-level computation.
- Fast BRAM access using a data-layout strategy that gives the inter- and intra-molecular energy modules 4× memory-access speed.
Results
On the same 140 ligand–receptor benchmarks, Vina-FPGA runs up to 6.9× (3.7× on average) faster than a state-of-the-art CPU while consuming only 2.5% of the energy, with similar docking accuracy. Compared with the GPU implementation Vina-GPU, its average energy consumption is just 45%.
Publications
- Ming Ling, Qingde Lin, Ruiqi Chen, Haimeng Qi, Mengru Lin, Yanxiang Zhu, Jiansheng Wu. “Vina-FPGA: A Hardware-Accelerated Molecular Docking Tool With Fixed-Point Quantization and Low-Level Parallelism.” IEEE Transactions on VLSI Systems, 31(4): 484–497, 2023. doi:10.1109/TVLSI.2022.3217275
- Follow-up work scaling the design across multiple boards: “Vina-FPGA-Cluster: Multi-FPGA Based Molecular Docking Tool With High-Accuracy and Multi-Level Parallelism.” IEEE Transactions on Biomedical Circuits and Systems, 2024. doi:10.1109/TBCAS.2024.3388323
- “Vina-FPGA2: a high-level parallelized hardware-accelerated molecular docking tool based on the inter-module pipeline.” Frontiers of Information Technology & Electronic Engineering, 2025. doi:10.1631/FITEE.2400941
