AntiBARTy Diffusion for Property Guided Antibody Design
This work addresses antibody design for therapeutic applications, offering an incremental improvement in in-silico methods for generating antibodies with specific properties.
The authors tackled the problem of designing novel antibodies with improved solubility by training an antibody-specific language model (AntiBARTy) and a property-conditional diffusion model, resulting in effective generation of novel antibodies with enhanced in-silico solubility while maintaining validity and controlling sequence diversity.
Over the past decade, antibodies have steadily grown in therapeutic importance thanks to their high specificity and low risk of adverse effects compared to other drug modalities. While traditional antibody discovery is primarily wet lab driven, the rapid improvement of ML-based generative modeling has made in-silico approaches an increasingly viable route for discovery and engineering. To this end, we train an antibody-specific language model, AntiBARTy, based on BART (Bidirectional and Auto-Regressive Transformer) and use its latent space to train a property-conditional diffusion model for guided IgG de novo design. As a test case, we show that we can effectively generate novel antibodies with improved in-silico solubility while maintaining antibody validity and controlling sequence diversity.