Skip to main content
Fig. 2 | Genome Medicine

Fig. 2

From: A robust deep learning workflow to predict CD8 + T-cell epitopes

Fig. 2

Effect of anchor and contact positions on peptide immunogenicity. A t-SNE embedding of peptides, coloured by HLA supertypes. Peptides at anchor positions (i.e. P2 and P9 of 9aa peptides) were represented by amino acid descriptors (aaDescriptors) and all amino acids across the peptide sequence were averaged to compute peptide-wide aaDescriptors. Each peptide was represented by positional (only at anchor positions) and peptide-wide aaDescriptors. B, C t-SNE embedding of peptide-wide descriptors and amino acids at contact positions, coloured by cognate TCRs for 9aa peptides (B) and 10aa peptides (C). D Sequence logos of peptides recognised by the same TCR, coloured by the physicochemical properties. Notably, these are examples of TCRs having the most cognate peptides in the database. However, except for the first TCR, the low number of peptides may limit statistical confidence in representing sequence conservation. E t-SNE embeddings of peptide-wide descriptors and amino acids at all positions, coloured by HLA supertype (left), immunogenicity (middle) and species of origin (right). F Distribution of intra-HLA and inter-HLA variation for peptides having different peptide-HLA entries. The variation described by ANOVA means squared

Back to article page