Major histocompatibility complex class I (MHC-I) molecules are principal components of immune recognition. Peptide binding to MHC-I molecules is necessary for antigen presentation and T cell recognition. Understanding and predicting MHC-I binding has great utility in T cell therapies, vaccine development and treatment of complex human diseases. Unfortunately, the polymorphic nature of MHC molecules makes universal prediction of peptide ligands challenging. High throughput peptidome characterization techniques enable experimental verification of possible MHC-I targets, but it is cost-prohibitive for broad clinical application. While much research has been conducted on predicting MHC-I-peptide binding, binding preferences of most MHC-I alleles remain unresolved, thus computational interventions are necessary.
Researchers at Arizona State University have developed a deep convolutional neural network, denoted HLA-Inception, for the prediction of MHC-I binding motifs. HLA-Inception integrates molecular electrostatic computations and pocket topology on the binding motifs, and is able to predict peptide binding motifs across 5,281 MHC-I alleles, covering most of the human population. The prediction algorithm derived from the generated motifs strongly correlates with quantitative peptide-MHC-I binding experiments which allows for predicting naturally presented MHC-I peptides with a precision which outperforms existing methods. This method can be used to predict the binding motif of MHC-I molecules which can then be used to predict peptide binding, analyze MHC-I associated diseases and develop therapeutics.
This model performs accurate and reliable motif predictions which are useful in MHC-I-peptide binding, but could also be applied to MHC-II molecules, protein-protein binding and TCR-MHC binding.
Potential Applications
-
Drug discovery
-
Development of broadly active vaccines
-
Development of immune epitopes
-
Developing T cell therapeutics
-
Analyzing and studying MHC-I associated diseases
-
Precision medicine
Benefits and Advantages
-
Improves generalizability
-
Faster and can be used for disease analysis in a way that is not realized with current analysis techniques
-
Higher precision and speed
-
The relatively simple mathematical operations enable a highly efficient network capable of quickly performing large proteome-scale predictions
-
Bridges the scales from molecular protein-peptide interactions to complex disease phenotypes
-
Training on electrostatic features detects the heterogeneity of binding motifs and improves universal peptide prediction
-
Offers accurate pan-allele MHC-I peptide ligand predictions
For more information about this opportunity, please see
For more information about the inventor(s) and their research, please see