Protein-based therapeutics may elicit undesired immune responses in a subset of patients, leading to the production of anti-drug antibodies (ADA). In some cases, ADAs have been reported to affect the pharmacokinetics, efficacy and/or safety of the drug. Accurate prediction of the ADA response can help drug developers identify the immunogenicity risk of the drug candidates, thereby allowing them to make the necessary modifications to mitigate the immunogenicity. In this study, we leveraged the rich clinical study data collected by Roche/Genentech to identify factors that impact drug immunogenicity. We focused on conventional monoclonal antibodies, but have included a variety of additional drug modalities in the analysis. We show that the clinical ADA incidences are associated with the mechanism of action of the drugs, the mechanism of action of comedications, the routes of drug administration and the diseases of the patient cohort. By combining these clinical factors with the in silico epitope prediction, we improved the prediction accuracy of drug immunogenicity in clinical trials (AUC of cross validation improved from 0.72 to 0.93).
Keywords: anti-drug antibody; clinical risk factors; immunogenicity; machine learning; protein-based therapeutics.
© 2025. The Author(s).