Benchmarking Peptide Spectral Library Search.

Area of Science:

• Proteomics
• Computational Biology
• Mass Spectrometry

Background:

• Spectral library search (SLS) is vital for peptide identification using tandem mass spectrometry.
• The accuracy of SLS heavily relies on Spectrum-Spectrum Matching (SSM) scoring functions.
• Existing comparative studies are hindered by a lack of comprehensive benchmark datasets.

Purpose of the Study:

• To introduce novel methods for constructing benchmarks for SSM scoring functions.
• To create a large-scale benchmark dataset for evaluating SSM scoring functions.
• To assess the performance of various SSM scoring functions and preprocessing methods.

Main Methods:

• Developed new methodologies for building SSM scoring function benchmarks.
• Constructed a benchmark dataset comprising eight query spectrum sets with varied noise levels (476,063 precursors).
• Included three spectral libraries (experimental, de-noised, predicted) with 3,065,819 precursors for evaluation.

Main Results:

• Identified significant limitations in current SSM scoring functions, with optimal recall around 70%.
• SpectraST performed best on low-noise spectra; JS-divergence demonstrated superior noise resistance.
• Cosine, Entropy, and Projected-Cosine scores performed poorly, especially with increasing noise.

Conclusions:

• The developed benchmark dataset (MSV000095946/PXD056205) facilitates the testing and development of new SSM scoring functions.
• The proposed benchmark construction approach offers a scalable framework for future SSM evaluation.
• Improving noise robustness in SSM scoring functions remains a critical challenge in proteomics.