Predicting antibody behavior in cation exchange chromatography (CIEX) remains challenging using conventional metrics like theoretical pI. We utilized high throughput analytical CIEX data from hundreds of antibody binding domains along with protein language model embeddings and sequence- and structure-derived descriptors to develop predictive ML models. These models provide improved tools for building blocks selection during multispecific antibody design and enable increased purification throughput by providing better prediction of CIEX separation.

Machine learning and generative AI are transforming antibody discovery by enabling prediction of vast antibody sequence spaces, increasing demand for scalable experimental validation. AstraZeneca has reengineered its automated infrastructure to support high-throughput generation of transfection-ready constructs. This presentation will describe the strategic development and deployment of AstraZeneca’s end-to-end automation platforms to reduce cycle times, expand library generation, and enhance production and delivery capacity for simple to complex antibody formats, enabling validation of data-driven predictions at discovery scale.