CARTnet: Template-guided B-cell Epitope Prediction with Contact-aware Attention

CARTnet, a contact-aware deep learning model for B-cell epitope prediction, reaches AUPRC 0.243 on the internal test set, exceeding DiscoTope-3 (0.220) and BepiPred-2 (0.140).

• Why it matters: B-cell epitope prediction is hindered by scarce experimentally labeled data and limited mechanistic understanding of antibody–antigen recognition, motivating more accurate and generalizable models.
• What they did: CARTnet combines multimodal inputs with template-derived epitope information and contact-aware attention, evaluated on internal and external test datasets using AUPRC; it also tests template-guided variants and cross-genus Alphavirus transfer.
• The result: On the external test set, CARTnet attains AUPRC 0.267 versus 0.192 (DiscoTope-3), 0.111 (BepiPred-2), and 0.215 (BepiPred-3), and a single template residue boosts AUPRC from 0.225 to 0.403 on the internal set.

AI summary of the abstract · may contain errors · not medical advice · read the paper.