Prediction of peptide binding to MHC using machine learning with sequence and structure-based feature sets.

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Peptide binding to Major Histocompatibility Complex (MHC) molecules is critical for adaptive immunity and vaccine design.
• Current machine learning models for peptide:MHC (p:MHC) binding prediction primarily rely on sequence data.
• Developing accurate predictive models is essential for identifying effective vaccine candidates against infections and tumors.

Purpose of the Study:

• To train and evaluate Support Vector Machine Classifier (SVMC) models for predicting peptide binding to the H-2D^b MHC I allele.
• To investigate the utility of physicochemical sequence-based and structure-based descriptors in p:MHC binding prediction.
• To identify key peptide features that influence binding affinity.

Main Methods:

• Trained SVMC models using physicochemical sequence-based and structure-based descriptor sets.
• Employed recursive feature elimination and two-way forward feature selection for model optimization.
• Evaluated model performance based on specificity, precision, and sensitivity.

Main Results:

• Models utilizing physicochemical descriptors achieved specificity and precision comparable to state-of-the-art sequence-based algorithms.
• The best-performing model was a hybrid approach combining both sequence-based and structure-based descriptors.
• Physicochemical properties of anchor positions (residues 5 and 9) were identified as significant contributors to binding affinity.

Conclusions:

• Machine learning models incorporating both sequence and structural data can effectively predict peptide:MHC binding.
• Physicochemical properties, particularly at anchor positions, play a crucial role in determining peptide binding affinity.
• These findings can inform the design of more effective vaccines by guiding peptide selection.