designing antibody sequencesat massive scale, predicting their properties, validating via experiments, and self-improving iteratively. We balance the tradeoff between exploration and exploitation and search the sequence landscape, aiming for more precise epitope targeting and better developability after each iterative design cycle
We train deep learning models to predict binding between antibodies and antigens. With ever-growing structural and functional data, our models become more precise and detailed.
Given an antigen, we efficiently search the antibody sequence space to identify molecules that target an epitope with significant therapeutic implications. A library of up to millions of antibodies are then synthesized for testing.
Using a variety of technologies, we obtain functional and biophysical measurements of individual antibodies in our library in a high-throughput manner.
By repeating the previous steps, we learn from the successes and failures of earlier experiments, train better models, design a new library of candidates, and further improve antibodies with desired properties.
