immuneML

immuneML implements each step of the adaptive immune receptor repertoire (AIRR) machine learning process in an extensible, open-source software ecosystem based on fully specified and shareable workflows.

Key Features:

• Exploratory Analysis: Comprehensive functionality for exploratory analysis of AIRR datasets.
• Machine Learning Training: Training machine learning models for repertoire or receptor sequence classification.
• Immune Event Simulation: Simulating immune events within AIRR data.
• Model Application: Applying trained machine learning models to new datasets with unknown class labels.
• Reproducibility and Interoperability: Fully specified and shareable workflows that address reproducibility, transparency, and interoperability in AIRR machine learning.
• Extensibility: Extensible architecture with developer documentation for extending functionality with new components.

Scientific Applications:

• Immune State Prediction: Reproduction of large-scale studies on immune state prediction from AIRR data.
• Antigen Specificity Prediction: Development, integration, and application of novel deep learning methods for predicting antigen specificity.
• Benchmarking: Streamlined interpretability-focused benchmarking of AIRR machine learning methods.
• Diagnostic and Therapeutic Discovery: AIRR-based diagnostic and therapeutic discovery through identification of complex discriminative sequence patterns.

Methodology:

immuneML implements machine learning workflows for AIRR analysis including training models for repertoire or receptor sequence classification, simulating immune events, and applying trained models to new datasets. The ecosystem supports deep learning methods for antigen specificity prediction and enables benchmarking of AIRR machine learning approaches through fully specified, shareable workflows.