Therapeutic antibody development requires precise co-optimization of myriad biophysical properties, many of which exhibit strict tradeoffs during protein engineering campaigns. One such tradeoff is often seen between arguably the two most important antibody properties: affinity and specificity. Co-optimization of these two properties remains challenging due to experimental limitations and obstacles associated with developing robust models that accurately predict analog binding values from deep sequencing datasets. We have sought to address these challenges by developing a comprehensive methodology for the acquisition of high-quality deep sequencing data and subsequent analysis with machine learning models. We have compared several sequence-based feature extraction methods for analysis via machine learning models, achieving accurate predictions of analog binding strengths from binary sequencing labels. This enables tunable Pareto optimization of antibody properties from high-throughput experimentation, which greatly reduces the amount of experimentation needed for co-optimizing key properties for generating potent antibody therapeutics.

Machine learning methods for Pareto optimal antibody engineering

ACS Fall 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.