Transformer-based de novo peptide sequencing for data-independent acquisition mass spectrometry

Area of Science:

• Proteomics
• Computational Biology
• Mass Spectrometry

Background:

• Tandem mass spectrometry (MS/MS) is crucial for high-throughput proteomics.
• Data-Independent Acquisition (DIA) offers non-targeted analysis but generates complex, multiplexed spectra.
• Current de novo peptide sequencing methods struggle with DIA data's multiplexing challenge.

Purpose of the Study:

• To introduce Casanovo-DIA, a deep-learning model for peptide sequencing from DIA-MS/MS data.
• To address the limitations of existing methods in handling multiplexed spectra.
• To improve the accuracy and efficiency of de novo peptide sequencing in proteomics.

Main Methods:

• Development of Casanovo-DIA, a deep-learning model utilizing transformer architecture.
• Application of the model to decipher peptide sequences directly from DIA mass spectrometry data.
• Benchmarking against state-of-the-art methods like DeepNovo-DIA and PepNet.

Main Results:

• Casanovo-DIA demonstrates significant improvements in precision and recall at both amino acid and peptide levels.
• Amino acid level precision increased by 15.14% to 34.8%, and recall by 11.62% to 31.94%.
• Peptide level precision saw a substantial boost of 59% to 81.36% compared to existing methods.

Conclusions:

• Casanovo-DIA effectively deciphers peptide sequences from challenging DIA mass spectrometry data.
• The model offers a promising approach for uncovering novel peptides and enhancing biological sample profiling.
• Integration of Casanovo-DIA with DIA data analysis advances the field of proteomics.