Deciphering the Language of Antibodies using Self-supervised Learning

An individual's B cell receptor (BCR) repertoire encodes information about past immune responses, and potential for future disease protection. One of the grand challenges of BCR sequence analysis is predicting BCR properties from their amino acid sequence alone, with over 1 billion BCRs in a single patient. We'll present an antibody-specific language model, AntiBERTa, trained on millions of BCR sequences on NVIDIA's GPUs. AntiBERTa captures biologically meaningful information about BCRs, making it generalizable for many downstream applications. Learn how AntiBERTa can be fine-tuned to predict key positions for molecule binding from an antibody sequence alone, outperforming public tools across multiple metrics. To our knowledge, AntiBERTa is the deepest protein family-specific language model; it provides a rich representation of BCRs that marks the first step toward a deep learning-guided understanding of the human immune system.