DeepTAPE: Enhancing Systemic Lupus Erythematosus Diagnosis with Deep Learning Based on TCRβ CDR3 Sequences

Systemic Lupus Erythematosus (SLE) is a common and severe autoimmune disease driven by abnormal T cell responses, which are critical for understanding the disease’s immune pathology. Given the central role of the Complementarity Determining Region 3 (CDR3) of the TCRβ chain in T cell specificity, focused investigation into CDR3 may potentially enhance both the diagnostic accuracy and the mechanistic understanding of SLE. In this study, we developed DeepTAPE, a deep learning-based engine for predicting autoimmune diseases. It primarily uses CDR3 sequence features, supplemented by the frequency distribution of TCRβ’s V genes. This model is based on a multi-layer CNN-LSTM architecture with residual connections. DeepTAPE exhibits superior diagnostic performance for SLE when compared to existing gene-focused studies. It achieves an impressive average AUC of 97.99%, accuracy of 93.97%, and recall of 94.36%, reaffirming the diagnostic utility of CDR3 in SLE through deep learning. Building on this, we introduced a quantitative indicator based on the model, the Autoimmune Risk Score, which is positively correlated with clinical disease activity in SLE patients and assists in prognosis. Overall, DeepTAPE not only offers accurate diagnosis but also enhances our understanding of SLE.