Binder Discrimination of Therapeutic Antibodies

Following antibody lead generation via conventional methods such as screening with phage display, selected candidates bind different epitopes of the target. However, antibodies that bind a specific epitope of interest are often required. With AI, we can select antibodies that bind a specified epitope within days, significantly lowering the number of candidates for further cost-intensive characterization.

Our test included 310 different Antibody-Antigen pairs, where 60 of them were approved therapeutic antibodies and 250 were non-therapeutic antibodies. We add to each antibody additional 9 variants that differ from the original one to emulate drug discovery screening conditions.

As a result, our AI solution can accurately distinguish antibodies that bind a specified epitope from ones that don’t. The predicted probabilities for binders are strongly skewed towards 1.0 with mean predicted probability of 95%. The predicted probabilities for binding of the non-binders are strongly skewed towards 0.0 with mean predicted probability of 18%.