Figure 4.

First principal component versus correlation based ranking plots of samples in biclusters identified by MCbiclust. (A) Fork patterns of the six biclusters found with MCbiclust in the synthetic data. Y axes show the first principal component (PC1) value for each sample in each bicluster. Principal component analysis was run on the most highly correlating samples and captures the correlation pattern present in the samples. X axes show the ordering according to how well the samples preserve this correlation. Ranking is obtained as described in the ‘Extending the bicluster – samples’ section of Supplementary Methods. (B) Mean centered average gene expression values of the two separate gene groups in the samples of the two forks of bicluster M1 determining the correlation. Expression levels in the two gene groups follow an antiparallel pattern. Relationship of average gene expression to PC1 values are shown in Supplementary Figure S5. (C) Schematics showing the gene-gene heatmaps of the M1 bicluster showing the division of the genes into two groups with different regulation in the upper and lower fork samples.