GPU-Accelerated Discovery of Pathogen Derived Molecular Mimics of a T-cell Insulin Epitope

Thomas Whalley, Gary Dolton, Paul E Brown, Aaron Wall, Linda Wooldridge, Hugo van den Burg, Anna Fuller, Jade R Hopkins, Michael D Crowther, Meriem Attaf, Robin R Knight, David K Cole, Mark Peakman, Andrew K Sewell, Barbara Szomolay

Research output: Contribution to journalArticle (Academic Journal)peer-review

54 Downloads (Pure)


The strong links between (Human Leukocyte Antigen) HLA, infection and autoimmunity combine to implicate T-cells as primary triggers of autoimmune disease (AD). T-cell crossreactivity between microbially-derived peptides and self-peptides has been shown to break tolerance and trigger AD in experimental animal models. Detailed examination of the potential for T-cell crossreactivity to trigger human AD will require means of predicting which peptides might be recognised by autoimmune T-cell receptors (TCRs). Recent developments in high throughput sequencing and bioinformatics mean that it is now possible to link individual TCRs to specific pathologies for the first time. Deconvolution of TCR function requires knowledge of TCR specificity. Positional Scanning Combinatorial Peptide Libraries (PS-CPLs) can be used to predict HLA-restriction and define antigenic peptides derived from self and pathogen proteins. In silico search of the known terrestrial proteome with a prediction algorithm that ranks potential antigens in order of recognition likelihood requires complex, large-scale computations over several days that are infeasible on a personal computer. We decreased the time required for peptide searching to under 30 min using multiple blocks on graphics processing units (GPUs). This time-efficient, cost-effective hardware accelerator was used to screen bacterial and fungal human pathogens for peptide sequences predicted to activate a T-cell clone, InsB4, that was isolated from a patient with type 1 diabetes and recognised the insulin B-derived epitope HLVEALYLV in the context of disease-risk allele HLA A*0201. InsB4 was shown to kill HLA A*0201+ human insulin producing β-cells demonstrating that T-cells with this specificity might contribute to disease. The GPU-accelerated algorithm and multispecies pathogen proteomic databases were validated to discover pathogen-derived peptide sequences that acted as super-agonists for the InsB4 T-cell clone. Peptide-MHC tetramer binding and surface plasmon resonance were used to confirm that the InsB4 TCR bound to the highest-ranked peptide agonists derived from infectious bacteria and fungi. Adoption of GPU-accelerated prediction of T-cell agonists has the capacity to revolutionise our understanding of AD by identifying potential targets for autoimmune T-cells. This approach has further potential for dissecting T-cell responses to infectious disease and cancer.
Original languageEnglish
Article number296
Number of pages18
JournalFrontiers in Immunology
Publication statusPublished - 28 Feb 2020


  • T-cell
  • Type 1 diabetes
  • T-cell receptor
  • peptide-HLA
  • Insulin
  • Molecular Mimicry
  • General-purpose computing on graphics processing units (GP-GPU)
  • Compute Unified Device 22 Architecture (CUDA)
  • Nvidia

Fingerprint Dive into the research topics of 'GPU-Accelerated Discovery of Pathogen Derived Molecular Mimics of a T-cell Insulin Epitope'. Together they form a unique fingerprint.

Cite this