SEMA

SEMA predicts conformational B-cell epitopes from antigen tertiary structures to support vaccine design and immunotherapeutic drug development.

Key Features:

• Deep transfer learning (ESM-1v, ESM-IF1): Fine-tunes pretrained models ESM-1v (protein language model) and ESM-IF1 (inverse folding model) to predict antibody–antigen interaction features.
• Quantitative residue-level prediction: Distinguishes epitope and non-epitope residues and provides quantitative assessments of immunodominance within antigen structures.
• Performance metrics: Demonstrates predictive performance with a ROC AUC of 0.76 on an independent test set.
• Application to SARS-CoV-2 RBD: Ranks immunodominant regions within the SARS-CoV-2 receptor-binding domain (RBD).

Scientific Applications:

• Vaccine development: Identifies potential conformational B-cell epitopes to inform rational vaccine antigen selection.
• Immunotherapeutic design: Supports design of antibody-based therapeutics by highlighting antibody-interacting regions on antigens.
• Infectious disease research: Prioritizes immunodominant regions in SARS-CoV-2 RBD for studies of neutralizing antibody responses.

Methodology:

Fine-tuning of pretrained models ESM-1v and ESM-IF1 to predict antigen–antibody interactions using primary sequence and tertiary structure information.