On graphics processing units for simulation of mass spectra

Abstract

Mass spectrometry (MS) is an important technique in analytical chemistry for identifying molecules. Accurate identification often relies on high-quality datasets of experimental mass spectra. Unfortunately, the existing datasets are not large enough to capture the full diversity of biological molecules. Expanding these datasets with new experiments is a slow and expensive process. An alternative to experimental data is the simulation of mass spectra in silico using computational quantum chemistry. These approaches are often computationally intensive, which limits their use to only the most well-equipped computational chemistry labs. In recent years, the mass adoption of Artificial intelligence (AI) programs has led to a soaring demand for high-performance graphics processing units (GPUs). This study demonstrates that offloading parts of the in silico mass spectra calculations to a GPU can lead to substantial speedups. In these settings, it is shown that a single state-of-the-art NVIDIA GPU can provide an equivalent computational power of over 300x central processing units (CPUs). In this way, it is demonstrated that the hardware typically used for AI inference can also be effectively utilized for the simulation of mass spectra.