Bacterial Immunogenicity Prediction by Machine Learning Methods.

Area of Science:

• Vaccinology
• Bioinformatics
• Machine Learning

Background:

• Identifying protective immunogens is crucial for vaccine design.
• Machine learning (ML) excels at analyzing large biological datasets like microbial proteomes.
• ML can significantly reduce experimental efforts in discovering vaccine candidates.

Purpose of the Study:

• To develop and validate machine learning models for predicting bacterial immunogenicity.
• To identify the most effective ML methods for distinguishing between bacterial immunogens and non-immunogens.

Main Methods:

• Applied six supervised ML methods: partial least squares-based discriminant analysis, k-nearest neighbor (kNN), random forest (RF), support vector machine (SVM), random subspace method (RSM), and extreme gradient boosting.
• Trained models on a dataset of 317 known bacterial immunogens and 317 non-immunogens.
• Validated models using internal cross-validation and an external test set.

Main Results:

• All tested ML models demonstrated good predictive ability for immunogenicity.
• The xgboost model showed the highest accuracy in identifying immunogens (84% on the test set).
• The combined RSM-kNN model achieved the best performance in recognizing non-immunogens (92% on the test set).

Conclusions:

• The study successfully developed accurate ML models for bacterial immunogenicity prediction.
• The top-performing models (xgboost, RSM-kNN, RF) have been integrated into the VaxiJen server.
• A majority voting approach using these models enhances the prediction of bacterial immunogens, streamlining vaccine development.