Label Noise Limits TCR-pMHC Specificity Prediction: Improved Performance Through AlphaFold3-Based Structural Modeling and Data Denoising

T cell receptor (TCR) binding to peptides presented by major histocompatibility complex (MHC) molecules is crucial for T cell activation and the foundation of adaptive immunity. Accurately predicting this specificity is vital for creating efficient TCR-based immunotherapies and vaccines. However, predicting TCR-pMHC specificity for new peptides remains a challenge, with structural modeling being the only strategy showing predictive potential thus far.

The study reveals that this limited performance is significantly influenced by label noise in the data used for training and evaluating these methods, a factor that has been largely overlooked until now. By utilizing an AlphaFold3-based pipeline specifically designed for TCR-pMHC structural modeling, researchers were able to achieve state-of-the-art specificity prediction, surpassing AlphaFold2.3-based and sequence-based methods.

Furthermore, when combined with a cluster-based denoising algorithm, removing mislabeled points from a large specificity dataset increased binder ranking accuracy by over 70% compared to the complete dataset. These findings emphasize label noise as a major obstacle limiting the performance of any method in this field, and demonstrate that integrating structural modeling with label denoising can substantially enhance TCR-pMHC specificity prediction.

This approach proves to be a promising complement to existing sequence-based methods for refining TCR target selection.