We have established a workflow integrating immunogenicity risk assessment with in silico analysis, which can trigger in vitro assays. This streamlined approach reduces the number of costly low-throughput in vitro tests and serves as a screening tool for selecting less immunogenic formats. We enhance initial immunogenicity risk assessments and develop effective mitigation strategies, safeguarding patient safety and improving therapeutic outcomes.

Understanding and predicting unwanted immunogenicity remains a central challenge in biotherapeutic development. This presentation will review key molecular, mechanistic, and clinical features contributing to anti-drug antibody risk, and provide practical guidance for locating and interpreting publicly available clinical immunogenicity and clinical-impact data. It will also highlight the Immunogenicity Database Collaborative (IDC), a grass-roots effort to standardize clinical immunogenicity datasets and accelerate development of interpretable, multivariable models that reflect the true clinical complexity of immunogenicity risk.

EpiVax has developed the ISPRI platform containing a multitude of tools for assessing 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, and leading to more accurate characterization of foreign and self epitopes within therapeutic molecules. New ADA models have led to a 3-fold increase in the correlation between predicted and observed ADAs over existing approaches, with over 75% of predicted ADAs within 10% of observed values.

Immunogenicity assessment is key for the development of biologics. Traditional approaches have focused on MHC class II antigen presentation. However, recent advances in B cell epitope and antibody-antigen interaction prediction have significantly enhanced predictive capabilities. This talk will describe some of these tools and introduces AbEpiTope-1.0, tool for predicting antibody targets, suggesting how these tools can be integrated into computational pipelines for immunogenicity assessment and de-risking of protein therapeutics.

We present a computational framework to predict TCR recognition of peptides presented by MHC-I and to design novel immunogenic peptides. Using HERMES, a model trained on the protein universe to predict amino acid preferences based on local structural environments, we accurately predict TCR–pMHC binding and T cell activity without task-specific training. We further design and experimentally validate de novo peptides that activate T cells and map peptide recognition landscapes across TCR–MHC systems.

Immunogenicity of biotherapeutics poses a significant efficacy or safety concern in drug development drug. It is crucial to select candidates with low immunogenicity risk or de-immunize the candidates with high risk at an early stage. In this presentation. I will introduce the tool of in silico prediction, domain competitive assay, and peptide screening for a multidomain therapeutics in preclinical and clinical studies. This integrated immunogenicity assessment will enhance the success ratio in drug development.

Advances in T cell epitope prediction have transformed computational immunogenicity assessment, motivating increased attention toward the complementary role of B cell epitopes. However, accurately modeling B cell recognition and incorporating conformational humanness evaluation into predictive frameworks remains a significant challenge. In this work, we present a prototypical workflow for leveraging B cell epitope prediction, together with structural humanness assessment, to enhance immunogenicity risk evaluation for biotherapeutics. Using large clinical datasets of anti-drug antibody responses, we show that integrating B cell epitope and humanness information improves both precision and recall in immunogenicity prediction compared to T cell epitope only approaches, highlighting its potential to refine preclinical risk assessment strategies.

Protein-based therapeutics can trigger anti-drug antibodies (ADAs) that affect pharmacokinetics, efficacy, or safety. Using Roche/Genentech clinical data, we identified factors influencing drug immunogenicity across monoclonal antibodies and other modalities. ADA incidence was linked to drug and comedication mechanisms, administration routes, and disease types. Combining these clinical factors with in silico epitope predictions improved the accuracy of clinical immunogenicity prediction.

This seminar presents quantitative systems pharmacology modeling of immunogenicity for biotherapeutic combinations, focusing on nivolumab and ipilimumab. The model showed that combining these agents increases nivolumab anti-drug antibodies compared to monotherapy but does not significantly alter pharmacokinetics, consistent with clinical data. This mechanistic approach to understanding immunogenicity can be applied to other biotherapeutic combinations.

Biotherapeutics may include motifs similar or identical to pathogenic sequences, acting as cross-reactive T cell epitopes that increase immunogenicity potential. IGMotifFinder is an in silico platform developed to identify HLA class II-presented cross-reactive motifs. Analysis of over 200 biotherapeutics revealed that a higher "pathogenic" to "self" motif-ratio correlates with increased clinical immunogenicity. IGMotifFinder will support early identification and proactive mitigation of immunogenicity in biotherapeutics.

Molecular sequence is a key contributor to immunogenic responses against biologic drugs. Leveraging in silico techniques enables early identification and mitigation of sequence-associated risks during the discovery phase, in a high-throughput and efficient manner. This presentation will highlight computational strategies that integrate a suite of in silico tools to optimize lead molecules for developability and reduce immunogenicity risk—ultimately accelerating the path to candidate selection.

In silico predictions of immunogenicity have long been a grand challenge in the design of polypeptide therapeutics. While a complete solution remains difficult, the power of novel de novo design methods can be coupled with ML methods to produce an automated, general solution for minimizing unwanted immunogenicity in de novo designed miniproteins. We present Immunovate, a lightweight solution for controlling miniprotein immunogenicity at several levels of simulation complexity.

HLAIIPred is an end-to-end and context-free deep learning model that predicts HLAII-presented peptides on the surface of antigen presenting cells. The model improves clinical immunogenicity prediction, identifies epitopes in therapeutic antibodies and prioritize neoantigens with high accuracy. In this presentation, the design methodologies integrated with HLAIIPred will be highlighted, providing important insights into peptide-HLAII interactions through the identification of core peptide residues. Additionally, screening strategies are discussed for predicting unwanted immunogenic segments in therapeutic antibodies using HLAII presentation models.