Antibody discovery using single-B cells, single-cell RNAseq, and NLP-driven language models

Exploring new monoclonal antibodies is crucial for applications such as immunotherapy or Diagnostics. Our laboratory employs several strategies to discover new monoclonal antibodies against viral pathogens such as SARS-CoV-2 and influenza A virus. Monoclonal antibodies against SARS-CoV-2 were isolated by analyzing single-cell RNAseq from RBD-specific B cells obtained from convalescent COVID-19 donors. After producing 44 antibodies in single-chain or full-IgG format, the antibody 19n01 demonstrated potent and broad neutralizing capacity in neutralization and competition assays against several variants, including Omicron BA.1, BA.2, and BA.4/5. While experimental screening remains the gold standard for identifying potent therapeutic antibodies, it is costly and time-consuming. Advances in high-throughput sequencing and experimental screening have generated large repertoires of characterized antibodies, creating new opportunities for computational modeling and candidate prioritization. We used AntiBERTy and evaluated multiple antibody input formats. Embeddings generated from the fine-tuned models were subsequently used as feature inputs to train additional machine learning classifiers. Unsupervised visualization revealed structured clustering patterns, and candidates were selected based on concordant predictions across deep learning and classical ML models. The prediction enables the identification of antibodies with broad binding capacity to recognize multiple SARS-CoV-2 variants. For influenza A, antibodies are isolated from single B cells sorted by flow cytometry, and the heavy- and light-chain sequences are determined. In all cases, antibodies are produced in Expi293.