Combining Three-Dimensional Modeling with Artificial Intelligence to Increase Specificity and Precision in...

Area of Science:

• Computational immunology
• Vaccine design
• Bioinformatics

Background:

• Accurate prediction of peptide-MHC binding affinity is vital for T cell-based vaccine development.
• Current sequence-based methods have limitations in predicting antigenicity.

Purpose of the Study:

• To develop a rapid, predictive computational approach combining sequence-based and structure-based methods.
• To improve the accuracy of predicting peptide-MHC binding affinity and reduce false positives.

Main Methods:

• Utilized NetMHCpan 4.0 (sequence-based artificial neural network) and 3D structural modeling (MODELLER, Rosetta FlexPepDock).
• Analyzed geometric variability of peptide conformations for strong vs. low-affinity binders.
• Applied thresholds for geometric fluctuations to enhance specificity.

Main Results:

• Structure-based approach significantly improved statistical specificity, reducing false positives.
• Combined method increased positive predictive value (PPV) for strong binders (Kd < 100 nM) from 40% to 52% (p=0.027).
• Average PPV increase of 10% across tested murine and human MHC alleles.

Conclusions:

• The combined computational approach offers a significant improvement over standalone sequence-based methods.
• This method aids in the rapid design of effective T cell-based vaccines by enhancing prediction accuracy.
• The approach is effective even for human MHC alleles with limited training data.