EpiVax is upgrading its computational immunology platform with AI-enabled models designed to give drug developers a more precise view of how the human immune system may respond to biotherapeutic candidates. The latest release of ISPRI combines an updated JanusMatrix model, a new tolerance-adjusted immunogenicity score and an enhanced anti-drug antibody prediction model, pushing computational assessment further upstream in biologic drug development.
Predicting whether a biologic drug will trigger an unwanted immune response is one of the difficult problems in biotherapeutic development. A candidate can perform well in laboratory studies and still provoke an immune reaction in patients, potentially reducing efficacy, creating safety concerns or forcing developers to redesign the molecule.
EpiVax is attempting to address that problem with a more AI-driven approach to in silico immunogenicity assessment.
The Rhode Island-based computational immunology company has released new AI-enabled capabilities for ISPRI, its platform for evaluating immunogenicity risk across biotherapeutic candidates and impurity sequences. The update brings together three major changes: JanusMatrix 2.1, a new JanusMatrix-adjusted EpiMatrix score called JAX, and ADA 2.2, an updated model for predicting anti-drug antibody responses.
The underlying idea is straightforward: rather than relying only on whether a therapeutic sequence contains potential T-cell epitopes, developers need to understand whether those sequences are likely to be recognized as foreign by the immune system and how human immune tolerance may modify that response.
That distinction becomes important for modern biologics, which can include engineered antibodies, recombinant proteins, vaccines and other complex therapeutic modalities.
Moving Beyond Simple Epitope Counting
EpiVax’s technology stack is built around computational analysis of immune epitopes — portions of proteins that can be recognized by immune cells.
Its EpiMatrix technology evaluates the potential for sequences to contain T-cell epitopes. JanusMatrix adds another layer by examining whether those epitopes have similarities to sequences found in humans, providing an indication of potential immune tolerance.
With JanusMatrix 2.1, EpiVax says it is applying AI to make that tolerance assessment more granular. The updated model weights human epitope cross-conservation using information about expression and prevalence.
In practical terms, that means a computational model is not treating every human-conserved sequence as equivalent. The biological context surrounding an epitope can influence how meaningful that similarity is when estimating immune tolerance.
That matters because immunogenicity is rarely a simple binary property. Two therapeutic sequences may contain potentially similar immune-recognition signals but produce different clinical outcomes depending on how those sequences interact with the human immune system.
The new JAX score builds on this concept by incorporating JanusMatrix-derived tolerance potential directly into EpiMatrix scoring. EpiVax describes the result as a more biologically informed measure of intrinsic effector epitope density.
For drug developers, the potential benefit is earlier visibility into immune liabilities before substantial resources are committed to clinical development.
Adding a Second Lens for Anti-Drug Antibodies
The third major component of the release is ADA 2.2, an updated model focused on anti-drug antibody prediction.
Anti-drug antibodies can develop when a patient’s immune system recognizes a therapeutic protein as foreign. Depending on the drug and the immune response, these antibodies can alter pharmacokinetics, reduce therapeutic activity or complicate treatment.
ADA 2.2 combines epitope-based measurements with biophysical characteristics and mode-of-action information. EpiVax says the model was trained using a diverse set of clinical monoclonal antibodies and shows stronger correlation between predicted and observed clinical immunogenicity.
The combination with JAX is potentially more significant than either model alone.
One model examines sequence-derived T-cell immunogenicity and tolerance, while the ADA model provides another perspective on antibody-mediated immunogenicity. Together, they can give developers multiple computational signals when assessing a candidate.
That is increasingly relevant as pharmaceutical companies seek to use AI not simply for discovering molecules, but for making earlier development decisions about which molecules should advance.
AI Moves Further Into the Biotech Development Stack
The EpiVax update also reflects a broader change in computational drug development. AI is moving beyond molecular discovery into areas such as safety assessment, biomarker analysis, clinical development and manufacturing.
Companies including Insilico Medicine, Recursion and Schrödinger have demonstrated different approaches to using computation and AI across the drug-development lifecycle. EpiVax’s focus is narrower: understanding how therapeutic sequences may interact with the human immune system.
That specialization could become increasingly important as biologic pipelines grow more complex.
The industry’s challenge is also shifting. Developers do not simply need models that can generate predictions; they need models that can connect those predictions to clinically meaningful outcomes and provide enough biological context for researchers to interpret the result.
EpiVax’s emphasis on tolerance, clinical antibody data and multiple complementary models reflects that direction.
The company also positions the release alongside the U.S. Food and Drug Administration’s increasing interest in New Approach Methodologies (NAMs) — approaches intended to provide more human-relevant evidence and potentially reduce reliance on traditional experimental methods in appropriate settings.
For biopharma companies, computational immunogenicity prediction is unlikely to eliminate laboratory or clinical validation. Its more immediate value is as a screening and decision-support layer that can help researchers identify problematic candidates earlier.
That could make AI-based immunogenicity assessment particularly useful in programs where developers are comparing multiple candidate sequences, optimizing therapeutic proteins or trying to understand why a promising biologic could generate an unwanted immune response.
The larger trend is clear: as AI becomes embedded across drug development, the competitive advantage may increasingly come from models that connect machine-learning predictions with validated biological knowledge.
For EpiVax, the latest ISPRI release represents a step in that direction — using AI not to replace immunology, but to make computational assessments of human immune response more context-aware and clinically relevant.
Market Landscape
The market for AI-driven drug discovery and computational immunology is expanding as pharmaceutical companies look for ways to reduce development risk before expensive clinical stages.
AI is increasingly being applied to target identification, molecular design, protein engineering and clinical development. The next layer involves predicting properties such as safety and immunogenicity earlier in the pipeline.
EpiVax competes in a specialized segment alongside broader computational drug-development platforms and protein-engineering technologies. Its differentiation is its focus on immune-response prediction and the use of human sequence conservation, epitope analysis and clinical data.
The broader opportunity is significant because biologics remain vulnerable to immune responses that can be difficult to predict using conventional approaches alone. As AI models become more capable of integrating sequence, structural and clinical information, computational immunogenicity could become a more routine component of candidate selection and optimization.
Top Insights
- EpiVax is using AI to refine predictions of immune tolerance and immunogenicity for biologic drug candidates before clinical development.
- JanusMatrix 2.1 adds expression and prevalence context to human epitope conservation, making tolerance predictions more biologically specific.
- The new JAX score combines tolerance potential with EpiMatrix measurements to produce a more integrated immunogenicity-risk assessment.
- ADA 2.2 uses clinical monoclonal-antibody data alongside epitope and biophysical features to improve anti-drug antibody prediction.
- The release reflects a broader shift toward human-relevant computational methods across the pharmaceutical development lifecycle.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
