The abstract is read at the publisher; the summary is JClub's.
CRAFT enables pMHC-conditioned de novo design of antigen-specific TCR CDR3β sequences under physicochemical constraints, achieving competitive naturalness with favorable binding and predicted structural plausibility.
- Why it matters: De novo antigen-specific TCR design is hindered by pMHC-dependent recognition complexity, strict interface physicochemical requirements, and limited paired training data. Existing generative methods largely rely on 1D sequence statistics and miss pMHC context, interfacial properties, and native receptor priors in one framework.
- What they did: They built CRAFT, a decoupled conditional generative model that reconstructs antigen fragments for TCRs, using PrimeBERT to encode pMHC context and PrimeTCR to decode CDR3β with native receptor priors. Cross-attention plus Atchley factor embeddings incorporate physicochemical constraints during antigen-conditioned sequence decoding, supported by large-scale pretraining and antigen-derived contextual guidance.
- The result: Across multiple evaluations, CRAFT shows competitive sequence naturalness, favorable performance in specific binding, and predicted structural plausibility, supporting efficient targeted candidate TCR discovery for personalized tumor immunotherapies.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.