An MCDM approach for Reverse vaccinology model to predict bacterial protective antigens

Area of Science:

• Bioinformatics
• Computational Biology
• Vaccine Development

Background:

• Reverse vaccinology (RV) utilizes computational methods for vaccine design.
• Machine learning (ML) improves RV accuracy, but challenges in prediction and accessibility remain.
• Accurate identification of bacterial protective antigens (BPAgs) is crucial for effective vaccine development.

Purpose of the Study:

• To develop a supervised ML-based method for classifying BPAgs.
• To identify consistently high-performing ML models for BPAg prediction.
• To propose a multi-criteria decision-making (MCDM) approach for selecting optimal ML models in RV.

Main Methods:

• Utilized six ML classifiers with physiochemical features from Protegen and Uniprot databases.
• Applied Synthetic Minority Oversampling Technique and Edited Nearest Neighbour (SMOTE-ENN) to address data imbalance.
• Implemented a soft and hard ranking model with the Technique for Order Preference by Similarity to the Ideal Solution (TOPSIS) and entropy for model selection.

Main Results:

• The proposed MCDM approach effectively ranks ML models for BPAg classification.
• Random Forest and Extreme Gradient Boosting were identified as top-performing models.
• The developed method demonstrated superior performance compared to existing open-source RV tools on benchmark datasets.

Conclusions:

• The novel ML and MCDM framework enhances BPAg identification accuracy in reverse vaccinology.
• This approach offers a more robust and accessible method for vaccine antigen prediction.
• The findings contribute to the advancement of rational vaccine design strategies.