Calibrated Abstention for Reliable TCR--pMHC Binding Prediction under Epitope Shift

TL;DR

This paper introduces a calibrated abstention framework for TCR--pMHC binding prediction, enabling models to abstain on uncertain cases for more reliable predictions, especially on unseen epitopes.

Contribution

It proposes a dual-encoder architecture with calibration and conformal abstention to improve prediction reliability under epitope shift.

Findings

Achieves AUROC 0.813 and ECE 0.043 on challenging splits.

Reduces ECE by 69.7% compared to uncalibrated baseline.

At 80% coverage, error rate drops from 18.7% to 10.9%.

Abstract

Predicting T-cell receptor (TCR)--peptide-MHC (pMHC) binding is central to vaccine design and T-cell therapy, yet deployed models frequently encounter epitopes unseen during training, causing silent overconfidence and unreliable prioritization. We address this by framing TCR--pMHC prediction as a \emph{selective prediction} problem: a calibrated model should either output a trustworthy confidence score or explicitly abstain. Concretely, we (1) introduce a dual-encoder architecture encoding both CDR3$\alpha$/CDR3$\beta$ and peptide sequences via a pre-trained protein language model; (2) apply temperature scaling to correct systematic probability miscalibration; and (3) impose a conformal abstention rule that provides finite-sample coverage guarantees at a user-specified target error rate. Evaluated under three split strategies -- random, epitope-held-out, and distance-aware -- our method achieves AUROC 0.813 and ECE 0.043 under the challenging epitope-held-out protocol, reducing ECE by 69.7\% relative to an uncalibrated baseline. At 80\% coverage, the selective model further reduces error rate from 18.7\% to 10.9\%, demonstrating that calibrated abstention enables principled coverage-risk trade-offs aligned with practical screening budgets.