A Framework for Database Search with AI Models in Mass Spectrometry-Based Proteomics.

Area of Science:

• Computational proteomics
• Bioinformatics
• Machine learning in mass spectrometry

Background:

• Database searching is the standard for peptide identification in mass spectrometry-based proteomics.
• Increasing dataset sizes and peptide search spaces strain computational resources.
• Machine learning (ML) is increasingly applied to spectrum identification and peptide-spectrum matching.

Purpose of the Study:

• To present emerging database search approaches for peptide detection.
• To develop a theoretical framework for analyzing runtime and scaling in spectrum identification.
• To compare classical search strategies with novel neural network-based methods.

Main Methods:

• Analysis of asymptotic complexity concerning the number of spectra and peptide candidates.
• Estimation of practical runtime and memory requirements on realistic hardware.
• Contrast of classical similarity functions with learned scoring models.

Main Results:

• A theoretical framework for evaluating computational performance of peptide search strategies.
• Identification of trade-offs between different search approaches.
• Estimation of resource needs for neural network-based versus classical methods.

Conclusions:

• The study provides a guide for selecting and developing scalable peptide search strategies.
• Highlights the potential for learned scoring models to augment or replace classical similarity functions.
• Addresses the computational demands of proteomics data in the era of large models.