Fig. 4

DI Algorithm performance. (a) The DI algorithm identifies more functionally related groups of genes than a random sampling of genes. Since functionally related genes tend to be regulated together this suggests that the DI algorithm is identifying genes that are differentially regulated between the wild-type and K13 mutant strains. At 6 h red blood cell invasion and exported proteins are over-represented amongst the down-regulated dephasing genes (GO:0098602--single organism cell adhesion, GO:0016337--single organismal cell-cell adhesion, GO:0007155--cell adhesion, GO:0048518--positive regulation of biological process,) and DNA replication and immune system processes are over-represented amongst the up-regulated dephasing genes (GO:0006260--DNA replication, GO:0006270--DNA replication initiation, GO:0006261--DNA-dependent DNA replication, GO:0002376--immune system process, GO:0002440--production of molecular mediator of immune response, GO:0002377--immunoglobulin production, GO:0006259--DNA metabolic process, GO:0044699--single-organism process). The graph for 24 h is Additional file 2: S1. (b) The DI algorithm consistently improves the correlations between 2 samples with each iteration while randomly removing genes does not. (c) The DI algorithm was tested on 100 simulated data sets and consistently identified the genes responsible for the poor correlations between the samples while randomly removing genes did not. The grey areas around the curves are the 95% confidence intervals. (d) Filtering out lowly expressed genes prevents the DI algorithm from identifying high variance-low confidence genes as dephasing (Additional file 2: S7). This volcano plot shows that the identification of dephasing driver genes is not biased by expression level like other methods of detecting transcriptome differences between samples