CapHLA: a comprehensive tool to predict peptide presentation and binding to HLA class I and class II

Area of Science:

• Immunology
• Computational Biology
• Oncology

Background:

• Human leukocyte antigen (HLA) proteins are crucial for immune response initiation via epitope presentation.
• Accurate prediction of peptide-HLA (pHLA) interactions is vital for effective cancer immunotherapies.
• Existing prediction tools are limited to either HLA class I or II, have peptide length constraints, and often lack sufficient accuracy.

Purpose of the Study:

• To develop a novel computational model, CapHLA, capable of predicting peptide presentation probability (PB) and binding affinities (BA) for both HLA class I and II.
• To integrate CapHLA predictions with antigen expression data to create a neoantigen quality model for predicting immunotherapy response.
• To evaluate the performance of CapHLA and the neoantigen quality model against existing methods and clinical data.

Main Methods:

• Development of CapHLA, a convolution and attention-based model trained on mass spectrometry data for pHLA binding and presentation prediction.
• Validation of CapHLA against 11 other prediction methods, focusing on HLA-II and non-classical peptide length datasets.
• Integration of CapHLA's PB and BA predictions with transcriptomic antigen expression (EP) data to build a neoantigen quality model.

Main Results:

• CapHLA demonstrated superior performance in predicting pHLA binding affinities and presentation probabilities compared to 11 other methods.
• The model showed particular strength in predicting for HLA-II and peptides of non-classical lengths.
• The neoantigen quality model, utilizing CapHLA predictions, significantly outperformed other genetics-based models in predicting immunotherapy response and patient prognosis in large cancer cohorts.

Conclusions:

• CapHLA is a versatile and accurate tool for predicting peptide presentation across both HLA class I and II.
• The developed neoantigen quality model shows significant prognostic value for predicting immunotherapy response and patient survival.
• This study provides a powerful new approach for neoantigen screening and enhances the potential of personalized cancer immunotherapy.