MoE-TCR

MoE-TCR: Mixture-of-Experts for TCR-Epitope Binding Prediction

This project presents MoE-TCR, a mixture-of-experts (MoE) framework for pan-specific T-cell receptor (TCR)–epitope binding prediction.

Overview

We develop a mixture-of-experts framework that leverages criss-cross attention mechanisms to improve the prediction of TCR–epitope binding interactions. The model is designed to handle pan-specific prediction tasks across diverse TCR and epitope sequences.

Key Features

• Mixture-of-Experts (MoE): Leverages multiple expert networks to capture diverse binding patterns
• Criss-cross attention: Enhances interaction modeling between TCR and epitope sequences
• Pan-specific prediction: Generalizes across different epitope specificities

Publication

This work has been published in Pattern Recognition:

Qu, Wei and Li, Jinxing and He, Zhentao and Wang, Jiayi and Zhu, Shanfeng. “MoE-TCR: Mixture-of-experts framework for pan-specific TCR-epitope binding prediction.” Pattern Recognition, 2026.

Read the paper

Keywords

Data mining, TCR epitope binding, MoE, Criss-cross attention

(Qu et al., 2026)