ParaEM: Antigen-aware sequence-based paratope prediction with Expectation-Maximization.

ParaEM achieves antigen-aware sequence-based paratope prediction with over 90% AUC-PR, rivaling structure-based methods without needing 3D structural data.

• Why it matters: Accurate identification of antibody paratopes is essential for understanding immune recognition and designing vaccines, but existing methods often rely on unavailable structural information.
• What they did: They developed ParaEM, which uses pretrained embeddings and a generalized EM framework to model antibody-antigen interactions directly from sequences, incorporating CDR and antigen context.
• The result: ParaEM outperforms existing sequence-based models and approaches structure-based accuracy, enabling scalable, antigen-aware antibody paratope prediction that enhances immunological research and therapeutic design.