Outliers in Figure 5 (upper middle part of the diagram), containing 'CyanoaminoOutliers in Figure 5

Outliers in Figure 5 (upper middle part of the diagram), containing “Cyanoamino
Outliers in Figure 5 (upper middle part of the diagram), containing “Cyanoamino acid metabolism”, “fatty acid biosynthesis”, “vitamin B6 metabolism” and “butirosin and neomycin biosynthesis”. These are likely false positives due to a limited number of SNPs in the genes of the respectiveBackes et al. BMC Genomics 2014, 15:622 http://www.biomedcentral.com/1471-2164/15/Page 7 ofFigure 4 Difference between row- and column permutations. The histograms in panel A and B show for two pathways the significance values as calculated for row and column permutations, respectively. Panels C and D present the BLU-554 side effects respective pathways as provided by KEGG. Here, red marked genes correspond to significant genes in our GWAS.pathways or a small number of participants in the pathway. To effectively adjust for such artifacts, the analysis could be restricted on larger pathways, however, leading to a loss of information on smaller paths. All significance values for enriched and depleted pathways are provided in Additional file 4: Table S3.Required number of permutation testsAnother important question in GWAS pathway analysis is how many permutations have to be carried out in order to obtain stable results with respect to the considered pathways? Here, one common choice is to generate 1,000 different permutations, just a small fraction of the exponentially growing permutation number. We explored the Coefficient of Variation (CV), the ratio of the standard deviation to the mean as potential criterion for estimating the required PubMed ID:http://www.ncbi.nlm.nih.gov/pubmed/25645579 number of permutations. In detail, we started by sampling 100 of the 20,000 permutation tests and stepwise increased the number. For each permutation set size 1,000 random drawings were carried out to calculate average value, standard deviation and CV value for column as well as row permutations.First, we considered the average and standard deviation for all pathways with 1,000, 2,000 and 5,000 permutation tests for row and column permutations separately. Additional file 5: Figure S3 shows exemplarily the dependency between column permutation test number and CV. Particularly for the significant pathways on the left of the vertical black line (p = 0.05), the difference between 2,000 permutations (blue) and 5,000 permutations (green) was not significantly larger than between 1,000 and 2,000 permutations. To exactly assess at which number of permutations the significance values converge for a certain pathway, PubMed ID:https://www.ncbi.nlm.nih.gov/pubmed/25447644 we estimated the influence of the number of column and row permutations on the significance for “pathways in cancer”. Figure 6 presents the average significance score and the respective standard deviation for up to 15,000 of these permutations in the upper panel. In the lower panel of that figure the coefficient of variation for both, column and row permutations, is presented. Here, it can be seen that significance values converge rapidly, resulting in our example in a moderate coefficient of variation, such that in our case indeed 2,000 permutations were sufficient to estimateBackes et al. BMC Genomics 2014, 15:622 http://www.biomedcentral.com/1471-2164/15/Page 8 ofFigure 5 Comparison between enriched and depleted pathways. Each dot corresponds to one pathway. Red dots correspond to depleted and green dots to enriched pathways.the actual significance value in a reasonably small confidence interval and coefficients of variation of approximately 0.1.Ulcerative Colitis (UC) pathwaysTo validate our approach, we evaluated GWAS data measured fro.