EpiVax launches AI immunogenicity models

EpiVax’s September 2 product launch is a useful window into where AI is actually showing up in biotech: not only in molecule generation, but in the quieter software layer that screens drug candidates for failure risk. The company said it added AI-enabled models to ISPRI, its in silico immunogenicity platform for biologic drugs, including an updated JanusMatrix model, a revised tolerance-adjusted score and a newer anti-drug antibody model. (prnewswire.com) That matters because immunogenicity is one of the recurring problems in biologics development. (prnewswire.com) A therapy can work as designed and still run into trouble if the immune system treats it as foreign and mounts a response, including anti-drug antibodies that can reduce efficacy or create safety issues. (epivax.com) EpiVax’s pitch is that better early prediction can help drug developers catch some of that risk before expensive lab and clinical work begins. ### What exactly did EpiVax add? EpiVax said the release centers on three model updates inside ISPRI. The first is JanusMatrix 2.1, which the company said applies AI to weight human epitope cross-conservation using expression and prevalence data. The second is a JanusMatrix-adjusted EpiMatrix score that EpiVax said is intended to align more closely with clinical immunogenicity outcomes. (prnewswire.com) The third is ADA 2.2, an anti-drug antibody prediction model that combines epitope-based measures with biophysical features and mode-of-action criteria. ISPRI itself is EpiVax’s licensed software platform for immunogenicity screening in drug discovery and preclinical development. (epivax.com) The company says the toolkit is used by large pharmaceutical companies and is designed to fit into preclinical R&D workflows rather than operate as a standalone academic model. ### Why is JanusMatrix a meaningful name in this announcement? JanusMatrix is one of EpiVax’s established algorithms for asking a specific immunology question: whether a candidate T-cell epitope resembles sequences in the human proteome closely enough to be seen as more “self-like” or tolerogenic. In EpiVax’s own technical materials and prior published descriptions, the model is used after epitope mapping to distinguish epitopes that may drive effector immune responses from those that may dampen them. (prnewswire.com) The September 2 update does not replace that logic so much as refine it. EpiVax said JanusMatrix 2.1 adds AI weighting around human cross-conservation, which the company said produces a more precise characterization of tolerance profiles. (epivax.com) In a 2025 company update, EpiVax had already said new models improved JanusMatrix precision and recall by 50%, giving some context for why the company is continuing to iterate on this part of the stack. ### Why does this sit in the “AI in biotech” bucket? Biotech companies have spent much of the past few years talking about AI for target discovery, protein design and lab automation. EpiVax’s announcement points to a narrower but commercially important category: decision-support software for development risk. (7157e75ac0509b6a8f5c-5b19c577d01b9ccfe75d2f9e4b17ab55.ssl.cf1.rackcdn.com) Instead of designing a drug from scratch, these models are meant to help rank, de-risk and potentially reengineer candidates before they move deeper into development. That makes the story partly about infrastructure. A platform like ISPRI depends on data pipelines, model maintenance, user interfaces and integration into regulated R&D workflows, which is where software engineering demand often appears in life-science toolchains. That inference is based on the company’s positioning of ISPRI as licensable workflow software rather than a one-off consulting output. (prnewswire.com) ### What should readers watch next? EpiVax said the updated models are part of ISPRI, which it offers through software licensing agreements to biotech and pharmaceutical companies. The next concrete signal will be whether the company publishes validation data, customer adoption details, or additional technical comparisons for JanusMatrix 2.1, the adjusted EpiMatrix score and ADA 2.2 in future product or conference materials. (epivax.com)