deepNGS navigator: exploring antibody NGS datasets using deep contrastive learning.

Area of Science:

• Immunology
• Bioinformatics
• Computational Biology

Background:

• High-throughput sequencing generates vast B-cell receptor (BCR) datasets, necessitating efficient computational tools for analysis.
• Identifying clonal families and related antibody sequences is crucial for understanding immune responses and antibody engineering.

Purpose of the Study:

• To introduce deepNGS Navigator, a novel computational tool for analyzing and visualizing large-scale BCR sequence data.
• To enable efficient detection and clustering of clonal families and related antibody sequences.

Main Methods:

• Leveraging language models and contrastive learning to transform antibody sequences into 2D representations.
• Utilizing sequence embeddings for mutational pattern inference and trajectory analysis.
• Developing a visualization tool for assessing dataset diversity and identifying sequence clusters.

Main Results:

• deepNGS Navigator generates intuitive 2D maps that visualize BCR sequence diversity and enable clustering based on sequence similarity.
• The tool facilitates the identification of clonal families and analysis of mutational patterns within projected sequence embeddings.
• Demonstrated utility across diverse datasets, including synthetic libraries, ML-generated data, immunized llama sequences, human B-cell sequences, and in silico clonal lineages.

Conclusions:

• deepNGS Navigator provides an effective method for exploring and analyzing large BCR sequence datasets.
• The tool aids in identifying potentially beneficial or problematic antibody sequences by visualizing sequence properties and relationships.