Adaptive GPU-accelerated force calculation for interactive rigid molecular docking using haptics

• A GPU data-parallel adaptive force calculation approach for haptic-based docking.
• Force updates in less than 2 ms for molecules comprising more than 100,000 atoms.
• Speed improvements of up to 90 times CPU-based force calculation approaches.
• No precomputation requirements on the receptor.
• A formula-based strategy for selecting at run-time the partitioning structure.

Molecular docking systems model and simulate in silico the interactions of intermolecular binding. Haptics-assisted docking enables the user to interact with the simulation via their sense of touch but a stringent time constraint on the computation of forces is imposed due to the sensitivity of the human haptic system. To simulate high fidelity smooth and stable feedback the haptic feedback loop should run at rates of 500 Hz to 1 kHz. We present an adaptive force calculation approach that can be executed in parallel on a wide range of Graphics Processing Units (GPUs) for interactive haptics-assisted docking with wider applicability to molecular simulations. Prior to the interactive session either a regular grid or an octree is selected according to the available GPU memory to determine the set of interatomic interactions within a cutoff distance. The total force is then calculated from this set. The approach can achieve force updates in less than 2 ms for molecular structures comprising hundreds of thousands of atoms each, with performance improvements of up to 90 times the speed of current CPU-based force calculation approaches used in interactive docking. Furthermore, it overcomes several computational limitations of previous approaches such as pre-computed force grids, and could potentially be used to model receptor flexibility at haptic refresh rates.

Figure optionsDownload high-quality image (298 K)Download as PowerPoint slide