Study of Sequence-based Deep Learning Models for Antibody Humanization on B-Cell Receptor Repertoire Sequencing Data

Antibody therapeutics are crucial for treating various diseases, and humanization is essential for ensuring their safety and efficacy. This study develops sequence-based deep learning models to humanize non-human-derived antibodies, enhancing their human likeness while minimizing laboratory efforts. Key components include identifying and manipulating residues to improve humanness and evaluating the risk of immunogenicity. The humanization tool built in this study is based entirely on B-cell receptor repertoire sequencing data as the training source, yielding results closely aligned with experimental findings. These advancements are expected to streamline antibody development and promise more efficient therapeutic solutions.