Koina

Koina democratizes machine learning for proteomics research by serving as an open-source model repository that enables remote execution of models via HTTP requests and integration into data analysis pipelines.

Key Features:

• Decentralized Model Repository: Koina facilitates publication and findability of machine learning models for proteomics as an open-source, online-accessible repository.
• Remote Model Execution: Koina enables usage of machine learning models via HTTP requests without requiring local installation or configuration of models.
• Pipeline Integration: Koina simplifies integration of machine learning models into proteomics data analysis pipelines, demonstrated with the FragPipe computational platform.
• Model Reusability: Koina addresses technical challenges in demonstrating reusability of published machine learning models for end-users in proteomics workflows.

Scientific Applications:

• Spectral Library Generation: Koina supports machine learning models that generate spectral libraries for proteomics experiments.
• Peptide Identification: Koina enables models that improve peptide identification in proteomics data analysis.
• Targeted Acquisition Optimization: Koina facilitates models that optimize targeted acquisition modes in mass spectrometry-based proteomics.

Methodology:

Koina uses an HTTP-based client-server architecture for remote model execution and integrates with existing proteomics software tools via HTTP requests.