S. Hochreiter

This paper demonstrates that deep generative models can design high-affinity, developable antibodies at scale using synthetic data and transfer learning.

The paper introduces Absolut!, a software suite generating synthetic antibody-antigen structures to benchmark machine learning methods for antibody specificity prediction.

The paper repurposes high-throughput imaging data from a glucocorticoid receptor assay to predict activity in unrelated drug discovery assays, significantly increasing hit rates and chemical diversity.