Fig. 3

Robustness of ICGEC. a Sensitivity of the ICGEC algorithm to random initial weights. ICGEC was run 11 times, one time with equal weights, and the other 10 times with randomly chosen initial weights. The correlation coefficients in regard to the gene scores (upper) and mark scores (lower) between the use of equal weights and random initial weights after 1 to 12 iterations are indicated. b Sensitivity of the ICGEC algorithm to a random subset of genes. ICGEC was run 11 times, one time with the full gene sets, and the other 10 times with approximately one-quarter randomly selected genes in the full gene sets. The correlations for gene scores (upper) and mark scores (lower) are indicated between the use of the full and partial data. It should be noted that the correlation coefficients were computed only for the sampled subset of genes. c Distributions of gene scores produced with original data (yellow) and permuted data (gray). d Mark scores produced with original data (blue) and permuted data (gray). The permuted data used in c and d were produced by breaking the row equivalence and column equivalence, respectively, of the original two ordered epigenetic circumstance matrices from H1 and MSC by shuffling. ICGEC was run 10 times to produce 10 different permuted datasets, from which the distributions of gene scores and marks scores were generated. The data in d are presented as the mean ± sd