ConvMHC

ConvMHC predicts peptide binding to Major Histocompatibility Complex (MHC) class I molecules using deep convolutional neural networks by encoding peptide–MHC interactions as image-like array (ILA) data for pan-specific prediction of HLA-A and HLA-B nonapeptides.

Key Features:

• Pan-Specific Prediction: Trained on nonapeptide HLA-A and HLA-B binding data encoded as ILA for predictions across multiple HLA alleles.
• Deep Convolutional Neural Networks (DCNN): Uses DCNNs to capture meaningful local patterns in ILA-encoded peptide–MHC interaction data.
• Locally-Clustered Interaction Detection: Detects locally-clustered interactions that synergistically stabilize peptide binding.
• Benchmark Performance: Reported F1 scores on IEDB benchmark datasets: HLA-A*31:01 = 0.86, HLA-A*03:01 = 0.94, HLA-A*68:01 = 0.67.

Scientific Applications:

• Peptide–MHC class I binding prediction: Prediction of nonapeptide binding to HLA-A and HLA-B alleles.
• Characterization of locally-clustered interaction patterns: Extension of methodology to protein/DNA, protein/RNA, and drug/protein interactions.
• Vaccine development and immunological studies: Informing computational vaccine design and related immunological research.

Methodology:

Peptide–MHC interactions are encoded as image-like array (ILA) data and analyzed using deep convolutional neural networks trained on nonapeptide HLA-A and HLA-B binding data and evaluated on IEDB benchmark datasets.