Fig. 9

SNP similarity analysis interpreting eQTL SNPs. a This toy example illustrates the SNP similarity analysis, which calculates pairwise semantic similarity between SNPs using the Experimental Factor Ontology (EFO). The input is a list of SNPs, with the option to include SNPs in linkage disequilibrium (LD). The output is a circos plot, with the link line colour graded according to the degree of semantic similarity between each pair of SNPs. The calculation of similarity takes into account the annotation profile of the SNPs, the information content (IC) of the term, and the term–term similarity. In our example, each SNP is directly annotated by two terms, and inherit additional annotation terms according to the true-path rule. The terms are coloured according to their IC; original terms have a rectangular border, inherited terms an elliptical border. SNP 1 shows similarity of varying degrees to the other three SNPs based on their shared annotation profiles. SNP 1 and SNP C share both “Term 1” and the very informative “Term 1.1.1.1”; as such, they have a very high degree of semantic similarity. SNP 1 and SNP A do not share any terms directly; however, SNP 1’s “Term 1.1.1.1” and SNP A’s “Term 1.1.1.2” are both child terms of “Term 1.1.1” and so a similarity measure can be calculated based on this term. “Term 1.1.1” is the most informative common ancestor (MICA) between the two SNP annotation profiles, meaning they have a relatively high degree of similarity. The MICA of SNP 1 and SNP B is “Term 1”. Since this term is less informative than the MICA of SNP 1 and SNP A (lower IC value), the similarity score between SNP 1 and SNP B is lower. b Semantic similarity results for real data. Global similarity output for cis-eQTLs induced by 24-h IFN-γ is shown in the circos plot (top left). The top similarity links involving a specific SNP, rs11150589, are shown in the main circos plot, together with DAG plots showing the terms annotating each SNP. The genes modulated by the eQTL SNPs are given in brackets