An interpretable open platform for sequence-based antibody developability prediction

Antibody developability, a crucial aspect of drug discovery, can now be predicted more accurately from sequence data thanks to a new open platform called DELPHI. Developed by researchers, DELPHI is designed to train developers to create these predictors using labeled antibody assay data. What sets DELPHI apart is its unique approach to reducing sequence-similarity leakage, ensuring more reliable results.

To demonstrate DELPHI's capabilities, researchers applied it to in-house polyreactivity and size-exclusion chromatography (SEC) data. The results were impressive, with mean AUC scores of 0.959 for polyreactivity and 0.933 for size exclusion. These scores indicate a high level of accuracy in predicting antibody developability.

Moreover, DELPHI showcased its potential for broader applications. Trained solely on the in-house data, the platform successfully transferred its predictive abilities to a public antibody library containing 246,293 entries. The AUC score of 0.950 for this public library demonstrates DELPHI's ability to rank polyreactivity at a level comparable to the best reported Ginkgo competition point estimate, all without using data from the specific competition.

DELPHI's versatility extends beyond prediction. It empowers laboratories to screen candidates before running expensive assays, generate residue-level engineering hypotheses, and retrain the platform for new assays. This flexibility makes DELPHI an invaluable tool for researchers in the field of antibody developability prediction, streamlining their work and enhancing the efficiency of their drug discovery processes.