Review: application and opportunities for machine learning and artificial intelligence in preclinical immunogenicity risk assessment

This review explores how artificial intelligence (AI), machine learning (ML), and quantitative systems pharmacology (QSP) are transforming immunogenicity risk assessment for biologic therapies. Immunogenicity — the unwanted immune response that generates anti-drug antibodies (ADAs) — remains a major challenge in biologics development because it can impact safety, efficacy, and clinical trial success.

The authors discuss how AI-driven prediction models, including machine learning algorithms for MHC Class II peptide binding and T-cell epitope identification, are being used to assess immunogenicity risk earlier in drug development. These computational approaches support candidate selection, protein engineering optimization, and preclinical risk assessment for biologics and next-generation therapeutics.

The paper highlights how advances in computational immunology, large biological datasets, and systems modeling are enabling more predictive and mechanistic approaches to drug development. However, the authors also emphasize ongoing industry challenges, including limited standardized datasets, inconsistent anti-drug antibody measurements, and difficulties validating predictive AI models across diverse biologic modalities.

Importantly, the review positions quantitative systems pharmacology (QSP) as a critical framework for integrating immune system biology, clinical observations, and mechanistic simulations to improve immunogenicity prediction and support model-informed drug development. The article also outlines future opportunities for AI/ML applications in clinical trial optimization, biologic design, and translational decision-making.

This publication is highly relevant for organizations developing monoclonal antibodies, protein therapeutics, vaccines, and other biologics seeking to reduce clinical risk, accelerate development timelines, and improve regulatory confidence through model-informed approaches.