AMULETY: A Python package to embed adaptive immune receptor sequences.

Area of Science:

• Immunoinformatics
• Computational Biology
• Bioinformatics

Background:

• Large language models (LLMs) are increasingly used to analyze adaptive immune receptors (AIRs).
• Diverse LLMs offer unique applications but pose challenges in accessibility and usability for researchers.
• A unified approach is needed to leverage various LLM strengths for AIR sequence analysis.

Purpose of the Study:

• To introduce AMULETY (Adaptive imMUne receptor Language model Embedding Tool), a Python package for generating AIR sequence embeddings.
• To enable users to utilize diverse pre-trained protein and antibody language models without complex configurations.
• To facilitate the application of LLM embeddings in downstream bioinformatics tasks.

Main Methods:

• AMULETY provides functions to generate embeddings for paired or single AIR sequences (heavy-light, alpha-beta, gamma-delta).
• The software supports pre-trained protein and antibody language models for sequence embedding.
• Embeddings are generated for amino acid sequences of adaptive immune receptors.

Main Results:

• Demonstrated variability in embedding spaces across different LLMs for antibody and T-cell receptor sequences.
• Showcased that distinct LLMs capture different aspects of epitope-binding characteristics.
• Validated AMULETY's utility in distinguishing between epitope-binding groups using AIR sequence data.

Conclusions:

• AMULETY enhances the accessibility and usability of LLM embeddings for adaptive immune receptor research.
• The tool allows researchers to explore and leverage diverse LLMs for analyzing AIR sequences.
• AMULETY supports various AIR chain configurations and pre-trained models, offering flexibility for immunoinformatics applications.