Immunogenicity risk is a critical consideration in the development of biotherapeutics, particularly during the lead selection phase. To enhance our ability to predict, manage, and mitigate the potential immunogenicity of therapeutic candidates, we have developed a robust strategy that integrates computational and in vitro tools for comprehensive risk assessment. This approach enables us to evaluate clinical tolerance risks and optimize drug candidates more effectively. In this case study, we demonstrate the practical application of these tools to assess immunogenicity risk for FLT3L-Fc. The study highlights how these methodologies can be employed synergistically to inform decision-making during therapeutic lead identification and optimization, ultimately contributing to the development of safer and more effective biotherapeutics.

We developed an integrated pipeline to predict and experimentally verify the immunogenic targets recognized by human T cells in viral family of potential pandemic concern. The approach is based on integration of published data curated in the IEDB, bioinformatic predictions and in vitro primary immunogenicity assay utilizing human T cells. The immunogenicity data is then integrated with sequence conservation across relevant phylogenetic spaces; and further integrated with AI-based immunogenic design.

Unwanted immunogenicity presents significant challenges to the safety and efficacy of biological drugs, and current computational and in vitro prediction tools have limited clinical relevance. Here, we introduce ImmunoScreen, an in silico tool for immunogenicity assessment, integrated within AstraZeneca’s lead selection and optimization workflows. We highlight novel approaches aimed at improving prediction accuracy, with a focus on identifying T cell epitopes that are cross-reactive with pathogen sequences. The utility of these methods is demonstrated through a case study comparing in silico, in vitro, and clinical outcomes for antibody panels. Additionally, we discuss approaches to predict clinical anti-drug antibody incidence by leveraging clinical data and incorporating factors such as mode of action, target and patient cohort characteristics.

Immunogenicity potential assessment should be started as early as possible in the biotherapeutic development process to inform de-immunization approaches and to avoid resources spending on candidates with a high inherent immunogenicity potential. Oftentimes, this is only possible using in silico tools. In my presentation, I would like to introduce the audience to the in silico-based workflow we implemented to analyze and decrease the immunogenicity potential of biotherapeutics in early development.

EpiVax has developed the ISPRI platform, containing a multitude of tools for assessing the immunogenic risk of biotherapeutics, including prediction of anti-drug antibody (ADA) responses. Novel AI/ML techniques have now been integrated into ISPRI, leading to improved performance. New models have enabled enhanced prediction of tolerated epitopes, improving both precision and recall of its JanusMatrix model by 50% and leading to more accurate characterization of epitopes within therapeutic molecules. In addition, new ADA models have led to a six-fold increase in the correlation between predicted and observed ADAs over existing approaches, with over 80% accurately predicted.

Evaluating the immunogenicity risk of T cell engagers poses distinct challenges, particularly due to the proliferative effects of the CD3-binding arm. This presentation outlines a comprehensive strategy that integrates in silico prediction algorithms, the MHC-associated peptide proteomics (MAPPs) assay, and a dendritic cell–PBMC co-culture proliferation model to characterize potential immunogenicity. A case study is presented to demonstrate how these tools collectively inform risk mitigation during the development of T cell engagers.

Understanding and predicting protein immunogenicity remains a central challenge in both therapeutic development and immunological research. We offer a fresh perspective by directly leveraging experimental immune response data combined with machine learning. This approach moves beyond traditional reliance on MHC binding predictions and on the Treg concept. This presentation will highlight the conceptual foundation and emerging applications of cytokine-informed models, emphasizing their potential as complementary tools in immunogenicity prediction.

This presentation discusses immunological principles to guide AI/ML epitope selection in cancer. I will cover the limitations of current vaccine trials and what we can do in the future. 

By analyzing real-world patient outcomes, adverse events, and biomarker responses, AI/ML can identify patterns and mechanistic insights that guide safer drug design. This approach improves predictive accuracy, reduces late-stage failures, and aligns preclinical testing with clinical realities, enabling more efficient development of therapeutics and facilitating targeted strategies for safety, efficacy, and personalized medicine.

LLMs and related chatbots have accelerated adoption of machine learning technologies, but fall far short in modeling the complexities of homeostatic human physiology or incorporating more human-curated approaches. RAG architectures are helpful, but full exploitation of neurosymbolic learning and agentic approaches in concert will drive the next generation of discovery.