Koina: Democratizing machine learning for proteomics research

Area of Science:

• Proteomics
• Bioinformatics
• Computational Biology

Background:

• Machine learning (ML) and deep learning show promise for proteomics applications like spectral library generation and peptide identification.
• Slow adoption of new ML models in proteomics is hindered by poor findability, accessibility, and integration challenges.
• Existing proteomics software often lacks straightforward methods for incorporating novel ML models.

Purpose of the Study:

• To present Koina, an open-source, decentralized, and online-accessible repository for ML models in proteomics.
• To facilitate the publication, discovery, and usability of ML models within the proteomics community.
• To demonstrate the seamless integration of ML models into existing proteomics data analysis workflows.

Main Methods:

• Development of Koina, an open-source, decentralized, and online-accessible platform for ML model repository.
• Implementation of an easy-to-use online interface for accessing and utilizing ML models.
• Integration of Koina with the FragPipe computational platform for proteomics data analysis.

Main Results:

• Koina provides a centralized solution for ML model findability and accessibility in proteomics.
• The platform enables straightforward integration of ML models into established data analysis pipelines.
• Demonstrated successful integration with FragPipe, showcasing improved proteomics data analysis capabilities.

Conclusions:

• Koina significantly lowers the barrier for adopting ML models in proteomics research.
• The repository fosters reproducibility and reusability of ML models for end-users.
• Koina represents a key advancement in computational proteomics, enabling broader application of ML.