EpitopeTransfer: a Phylogeny-aware Transfer Learning Framework for Taxon-specific Linear B-cell Epitope Prediction

Area of Science:

• Immunoinformatics
• Computational Biology
• Vaccine Development

Background:

• Accurate prediction of linear B-cell epitopes (LBCEs) is crucial for developing immunodiagnostics, vaccines, and therapeutics.
• Current computational predictors often require extensive pathogen data, leading to biases and reduced performance on understudied or emerging pathogens.
• Generalist models trained on broad datasets may not capture specific pathogen characteristics effectively.

Purpose of the Study:

• To introduce EpitopeTransfer, a novel phylogeny-aware transfer learning strategy for enhanced LBCE prediction.
• To address the limitations of existing LBCE prediction methods, particularly their performance on neglected or emerging pathogens.
• To improve the accuracy and generalizability of computational epitope prediction models.

Main Methods:

• Developed EpitopeTransfer, a transfer learning approach utilizing large protein language models.
• Fine-tuned models with abundant data from higher-level taxa (e.g., broader taxonomic groups).
• Applied refined feature embeddings to train pathogen- or lower taxon-specific LBCE prediction models.

Main Results:

• EpitopeTransfer demonstrated substantially increased predictive performance across viruses, bacteria, and eukaryotes compared to state-of-the-art methods.
• The gains in performance were attributed to the phylogeny-aware fine-tuning of the feature embedder and taxon-specific model optimization.
• The strategy effectively overcomes biases associated with generalist models and improves prediction for diverse pathogens.

Conclusions:

• Phylogeny-aware transfer learning significantly enhances the accuracy of linear B-cell epitope prediction.
• EpitopeTransfer offers a robust solution for predicting LBCEs, especially for underrepresented pathogens.
• This approach accelerates the discovery and prioritization of immunological targets for vaccine and therapeutic development.