An Ultrafast GPU-Enabled MGVB.

Background: cMGVB is a graphical processing unit (GPU)-enabled implementation of the computational proteomics data analysis toolset MGVB. MGVB was released in 2025 as a Linux program designed to run on multi-node servers. It utilizes a novel algorithm for finding combinations of post-translational modification in peptide MS/MS data. The original combinatorial algorithm required a significant amount of resources to be practical. Hence, the aim of the research reported here was to port the algorithm to GPU and thus increase its speed and efficiency. Methods: To accomplish this it was recoded in CUDA C; recursive functions and data structures were re-implemented as non-recursive, and the algorithm was incorporated in a new version of MGVB, now termed cMGVB. Results: The re-implemented algorithm is much faster and, unlike the original program, can run on single CPU workstations equipped with inexpensive GPUs and still be much faster than the original algorithm running on HPC clusters. A typical focused search is completed in about a minute by cMGVB compared to 10-15 min by the original implementation. Illustrative case studies are presented and discussed in this report. Conclusions: cMGVB enables workflows that were not practical or even possible with the original MGVB.