MSSort-DIA-XMBD

MSSort-DIA-XMBD classifies peptide precursors quantified by OpenSWATH from data-independent acquisition (DIA) mass spectrometry using deep learning on extracted ion chromatograms (XICs) to distinguish high-confidence from low-confidence precursors.

Key Features:

• Deep convolutional neural network (CNN): Implements a CNN architecture trained on approximately 50,000 manually classified peptide precursors.
• XIC-based pattern recognition: Uses extracted ion chromatograms (XICs) as input and recognizes signal patterns associated with precursor identification confidence.
• Double-threshold segmentation: Applies a double-threshold segmentation strategy that reduces the proportion of low-confidence precursors requiring manual inspection to below 10%.
• Processing throughput: Provides automated classification throughput on the order of ~20,000 peptide precursors in minutes.
• OpenSWATH integration: Operates on OpenSWATH output and further discriminates precursors that remain after OpenSWATH's 1% false discovery rate control, including those with low fragment similarity.
• Cross-platform training: Model training includes data from various instrument platforms and species.

Scientific Applications:

• Proteomics quantification: Improves reliability and efficiency of peptide quantification in proteomic studies by distinguishing high-confidence precursors.
• OpenSWATH DIA result curation: Reduces manual inspection burden for OpenSWATH DIA reports by filtering low-confidence precursor candidates.
• Cross-platform DIA analysis: Applies to DIA datasets from multiple instrument platforms and species due to diverse training data.

Methodology:

Uses a deep convolutional neural network trained on ~50,000 manually classified peptide precursors with extracted ion chromatograms (XICs) as input, combined with a double-threshold segmentation strategy to filter low-confidence precursors.