EMBER: Predicting TCR--pMHC Binding Without a Docked Complex

Hugging Face Daily Papers 10/19/26, 12:00 AM Papers

protein-binding immunology structural-biology graph-representations se3-invariance tcr-pmhc deep-learning

Summary

EMBER 提出一种无需对接复合物结构的 TCR–pMHC 结合预测框架，将 TCR 与 pMHC 独立折叠并编码为 SE(3) 不变残基图进行比较，避免了对接误差的传播，同时降低计算复杂度，并在表位不相交评估中全面超越六个基线模型。

We propose EMBER, a structure-based framework for TCR--pMHC binding prediction that eliminates the need for a predicted docked complex, and with it the placement error that dominates such predictions. Traditional structure-based predictors utilize the co-folded complex structure of the receptor and the antigen and interpret binding from its contact surface. Consequently, every downstream feature inherits any deviation the structure-prediction tool introduced. Methodologically, EMBER folds the TCR and pMHC independently, encodes them as SE(3)-invariant residue graphs, and compares their learned representations without constructing a shared coordinate frame. A relative pose is therefore not an input of the model, so placement error cannot propagate into the prediction. Computationally, EMBER preprocesses O(N{+}M) monomer folds where a complex-based predictor needs O(NM) operations. Experimentally, under epitope-disjoint evaluation with six retrained baselines, EMBER is first on every metric under all four folding tools, by 0.9 to 14.4% in micro-AUROC, 18.5 to 25.3% in micro-AUPRC and 10.6 to 24.1% in MCC. Its ranking is also up to 85% more stable across folding tools.

Original Article

View Cached Full Text

Cached at: 10/21/26, 07:03 AM

Paper page - EMBER: Predicting TCR–pMHC Binding Without a Docked Complex

Source: https://huggingface.co/papers/2610.27334

Abstract

EMBER predicts TCR-pMHC binding from independently folded monomers using SE(3)-invariant graphs, avoiding docking errors and reducing computation while improving accuracy across folding tools.

We propose EMBER, astructure-based frameworkfor TCR--pMHC binding prediction that eliminates the need for a predicteddocked complex, and with it the placement error that dominates such predictions. Traditional structure-based predictors utilize theco-folded complexstructure of the receptor and the antigen and interpret binding from itscontact surface. Consequently, every downstream feature inherits any deviation the structure-prediction tool introduced. Methodologically, EMBER folds the TCR and pMHC independently, encodes them asSE(3)-invariant residue graphs, and compares their learned representations without constructing ashared coordinate frame. Arelative poseis therefore not an input of the model, so placement error cannot propagate into the prediction. Computationally, EMBER preprocesses O(N{+}M)monomer foldswhere a complex-based predictor needs O(NM) operations. Experimentally, underepitope-disjoint evaluationwith six retrained baselines, EMBER is first on every metric under all four folding tools, by 0.9 to 14.4% inmicro-AUROC, 18.5 to 25.3% inmicro-AUPRCand 10.6 to 24.1% inMCC. Its ranking is also up to 85% more stable across folding tools.

View arXiv page View PDF Project page Add to collection

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2610.27334 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2610.27334 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2610.27334 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Mirror-Score: Calibrated, Inference-only Scoring Exposes the Limits of Sequence-compatibility Ranking in D-peptide Design

arXiv cs.AI

Mirror-Score introduces a calibrated, inference-only scoring framework for D-peptide/L-protein binder design, releasing the first public benchmark of 31 crystal complexes and showing that raw ProteinMPNN NLL is a poor affinity ranker (Spearman 0.19), while Boltz-2 interface pLDDT achieves structure-level leave-one-out Spearman 0.90 on the viral-entry family. The paper also reports that cross-family calibration fails to transfer and proposes a prospective design protocol for antimicrobial-resistance targets LasR and LecB.

ConTact: Contact-First Antibody CDR Design via Explicit Interface Reasoning

arXiv cs.LG

ConTact introduces a contact-then-act architecture for antibody CDR design that explicitly decomposes the task into interface reasoning, contact prediction, and contact-gated sequence generation, achieving state-of-the-art structural quality and epitope awareness on the Chimera-Bench benchmark.