AntiRef: reference clusters of human antibody sequences.

Area of Science:

• Immunoinformatics
• Computational Biology
• Bioinformatics

Background:

• Human antibody sequence datasets suffer from genetic biases, leading to numerous duplicate and highly similar sequences.
• Existing datasets are skewed towards specific diseases like cancer and certain infections (HIV, influenza, SARS-CoV-2), limiting broader applicability.
• Data redundancy and biases impede efficient similarity searches and the training of statistical or machine-learning models.

Purpose of the Study:

• To address the computational challenges of clustering large antibody sequence datasets.
• To develop optimized clustering thresholds for antibody sequences, improving upon general protein clustering methods.
• To create publicly available, clustered antibody sequence datasets for broader research use.

Main Methods:

• Developed Antibody Reference Clusters (AntiRef), a system modeled after UniRef, specifically for antibody sequences.
• Applied antibody-optimized identity thresholds to cluster a large dataset of approximately 451 million full-length, productive human antibody sequences.
• Generated reference datasets clustered at various identity thresholds, including AntiRef90 and AntiRef100.

Main Results:

• AntiRef provides clustered datasets of filtered human antibody sequences using optimized thresholds.
• AntiRef90 is significantly smaller than the input dataset (one-third the size) and the non-redundant AntiRef100.
• The clustering approach effectively reduces redundancy and improves data usability for downstream analyses.

Conclusions:

• AntiRef offers a computationally efficient solution for managing and analyzing large-scale human antibody sequence data.
• The optimized clustering provides valuable, non-redundant datasets for researchers, facilitating antibody discovery and engineering.
• Accessible datasets and code promote wider adoption and advancement in antibody-related research.