HOMS-TC

HOMS-TC accelerates spectral library searching in mass spectrometry (MS) proteomics by using hyperdimensional computing and NVIDIA tensor cores to enable open modification searching (OMS) for identification of peptides with unexpected post‑translational modifications (PTMs).

Key Features:

• Open modification searching (OMS): Implements OMS to annotate modified peptides by finding partial matches with their unmodified counterparts in spectral libraries.
• Hyperdimensional encoding: Encodes mass spectral data into hypervectors using hyperdimensional computing principles, minimizing information loss and allowing independent calculation of each dimension for parallelization.
• Full-pipeline parallelism: Exploits parallelism across the entire spectral library search pipeline to accelerate computation.
• Cascade search parallelization: Processes two stages of the existing cascade search in parallel to efficiently select the most similar spectra while accounting for PTMs.
• NVIDIA tensor core acceleration: Offloads computation to NVIDIA tensor cores in modern GPUs to increase throughput.
• Performance: Empirical evaluations report approximately 31× speedup relative to alternative search engines while maintaining comparable identification accuracy.
• High-dimensional data handling: Optimized for efficient and accurate processing of high-dimensional MS spectral representations.

Scientific Applications:

• Spectral annotation: Annotating experimental mass spectra in MS proteomics, including spectra from peptides with unexpected PTMs.
• Modified peptide discovery: Identifying peptides with novel or unanticipated post‑translational modifications by matching to library spectra via OMS.
• Large-scale proteomics: Enabling scalable spectral library searches for growing MS proteomics datasets through accelerated runtimes.
• Cascade search workflows: Improving cascade search workflows to select most similar spectra while accounting for modifications.

Methodology:

HOMS-TC encodes mass spectra into hypervectors via hyperdimensional computing, applies open modification searching to find partial matches to unmodified library spectra, executes two cascade search stages in parallel, and accelerates computation on NVIDIA tensor cores in modern GPUs.