An Out-of-Core GPU based dimensionality reduction algorithm for Big Mass Spectrometry Data and its application in...

Area of Science:

• Computational Biology
• Biotechnology
• Bioinformatics

Background:

• High-resolution Mass Spectrometry (MS) generates millions of spectra in systems biology.
• Pre-processing MS data to remove non-useful peaks is crucial but challenging due to algorithm time-complexity.
• Existing sequential noise-reduction algorithms are often impractical for large-scale MS data analysis.

Purpose of the Study:

• To present a GPU-based dimensionality-reduction algorithm, G-MSR, for efficient MS2 spectra pre-processing.
• To introduce novel data structures, Binary Spectra and Quantized Indexed Spectra (QIS), for optimized GPU computation.
• To address GPU memory limitations by implementing adaptive in-core and out-of-core processing modes.

Main Methods:

• Developed a GPU-based algorithm (G-MSR) for MS2 spectra dimensionality reduction.
• Utilized novel data structures: Binary Spectra for efficient CPU-GPU communication and QIS for 3D to 1D data transformation.
• Implemented dynamic switching between in-core and out-of-core processing to manage GPU memory constraints.

Main Results:

• G-MSR achieves a peak speed-up of 386x compared to sequential algorithms.
• The algorithm successfully processed over one million spectra in just 32 seconds.
• Novel data structures optimized memory usage and computational efficiency on GPUs.

Conclusions:

• G-MSR offers a highly efficient solution for pre-processing large-scale MS2 spectral data.
• The algorithm's speed and memory optimization make it suitable for complex systems biology studies.
• The open-source availability of G-MSR facilitates its adoption in the research community.