Tesorai Search: cloud-based database search engine boosts identifications for mass spectrometry proteomics with a...

Area of Science:

• Proteomics
• Computational Biology
• Mass Spectrometry

Background:

• Traditional mass spectrometry search engines use basic algorithms for peptide identification.
• Recent advancements incorporate complex components like fragment ion intensities and retention time predictions, sometimes yielding inconsistent results.
• Decoy-based strategies are common but can complicate analysis.

Purpose of the Study:

• To evaluate a novel deep-learning approach for peptide identification that bypasses the need for decoys.
• To assess the impact of a simplified workflow using a pretrained deep-learning model on mass spectrometry data analysis.
• To compare the performance of the new method against established software.

Main Methods:

• A deep-learning pretrained model was developed to learn direct relationships between full spectra and peptide sequences.
• This model replaces traditional extra components like fragment ion intensity prediction and decoy-based classifiers.
• The simplified workflow was tested across diverse datasets and instrument types.

Main Results:

• The deep-learning approach consistently identified more peptides compared to FragPipe (12% more), PEAKS (9% more), and Proteome Discoverer (21% more).
• The simplified workflow demonstrated robustness on data from instruments and use-cases not encountered during training.
• Tesorai Search exhibited high speed, processing 250 immunopeptidomics searches in 45 minutes.

Conclusions:

• Replacing complex components with a single deep-learning model offers a simplified, robust, and more accurate peptide identification strategy.
• The Tesorai Search method provides significant improvements in peptide identification rates across various datasets.
• This approach facilitates easier use and more reliable performance in mass spectrometry-based proteomics.