FPGA Acceleration of Rigid Molecule Interactions

Abstract

Abstract. Modeling of molecule interactions often uses two or more successive models of increasing complexity. Rigid models based on correlation techniques are common as early screening passes-to detect interactions worth costlier examination-and are often at the heart of later passes as well. Even these rigid models are time-consuming when applied to large models at 10 3 − 10 5 different three-axis rotations. This paper presents an FPGA structure for performing the correlations efficiently by using a systolic array for 3-D correlation and arithmetic tailored to the application. The system includes a novel addressing technique for performing a three-axis rotation of a 3-D voxel model using modest amounts of logic and nearly no cost in time or buffer space. We compare our FPGA implementation with one on a PC using the standard transform-based method and find a speed-up of a factor of 200. We present extensions for handling implementation technologies with different performance characteristics and for handling models too large to fit on-chip.

GPU optimizations for a production molecular docking code

“…As each evaluation can take many CPU-hours, huge processing capability must be applied. Docking codes have been accelerated with FPGAs [15, 18, 20, 21], Cell [14], and GPUs [13, 17, 19]. …”

FPGA acceleration of rigid-molecule docking codes

“…Our GPU work accelerates the PIPER docking code [5]; we summarise this below and compare it with the FPGA approach presented here. The present authors and collaborators have also previously used FPGAs to demonstrate proof-of-concept FPGA-based acceleration of ZDOCK and some other systems [6, 7]. …”

“…In our previous work [6, 7] we showed that, for FPGA-based coprocessors, the original direct correlation – rather than an FFT – is sometimes the preferred method for computing rigid molecule docking. Two reasons for this are the inherent efficiency with which FPGAs perform convolutions and the modest precision (2–7 bits) of the original voxel data.…”

Hardware Accelerated Molecular Docking: A Survey

“…This implementation is described in references [21,22]. The applied algorithm uses 3D correlation which is a common rigid-body docking technique.…”

“…In order to exploit this the implementation has a flexible structure; the design can be easily configured to adapt different scoring schemes. The initial implementation [21] used a very simple voxel type consisting of only two bits that distinguish molecule interiors from exteriors and mark the surface of the molecules. The final version allows using voxels with tuple data type that represent different effects including directional interactions like hydrogen bonding.…”