GPU-accelerated algorithm for fast computation of biomolecular isotopic envelopes

Abstract

To accommodate the new features of modern protein mass spectra with Nobel-prize-winner electrospray ionization, Zhixin Tian, et al. developed isotopic Mass-to-charge ratio and Envelope Fingerprinting (iMF) algorithm for in situ interpretation and database search of protein tandem mass spectra. The creation of the customized theoretical database of both proteins and their dissociation fragment ions requires efficient computation of isotopic envelopes. This paper presents a GPU-accelerated algorithm for rapid computation of isotopic envelopes on NVIDIA Compute Unified Device Architecture (CUDA) platform, which can achieve 17.6 speedup when computing isotopic distributions of 512 fragment ions with mass 27 kda on average on IBM Power 8 with NVIDIA Tesla K40 m. Through optimizations on both CUDA memory access and stream scheduling, we find out a proper solution with chunk size = 32 in each kernel and double buffer cache can just hide the latency of memory and solution with chunk size = 64 can achieve about 160 speedup on the same experiment environment. The experimental results show that parallel algorithm with GPU and multiple optimization strategies provide an effective method with high performance to calculate isotopic envelopes.