Dual threshold optimization (DTO) and network inference in yeast. (A) Numbers of acceptable TFs, unique target genes, and network edges, when comparing Harbison ChIP data to TFKO or ZEV15 response data. “Unique targets” are genes that are in the bound-responsive intersection of at least one acceptable TF and thus are plausible direct functional targets. Edges connect acceptable TFs to the genes in their bound-responsive intersection. The ZEV15 response data yield more acceptable TFs, more unique targets, and more regulatory edges. (B) Illustration of DTO algorithm. Each dot represents one gene. Red lines indicate the chosen (optimal) thresholds for binding (vertical red line) and regulation (horizontal red line). The lower left quadrant, relative to the red lines, contains the bound and responsive genes, which are presumed to be direct functional targets (red dots). Gray lines indicate some of the other possible thresholds on binding or response, and locations where the gray lines cross are possible combinations of binding and response thresholds, each of which is evaluated by the DTO algorithm. (C) Numbers of acceptable TFs and unique target genes for comparison of Harbison ChIP binding data to TFKO or ZEV15 response data, after dual threshold optimization. The requirement that the overlap between the bound and responsive targets be greater than chance at P < 0.01 was checked by comparing the nominal hypergeometric P-value for the overlap to a null distribution obtained by running dual threshold optimization on 1000 randomly permuted binding and response data sets. DTO increases the network size, relative to using fixed significance thresholds. ZEV15 still yields more acceptable TFs, regulated genes, and regulatory interactions than TFKO. (D) Comparison of TFKO and ZEV15 networks derived from fixed thresholds, DTO on raw gene expression, and DTO on gene expression data processed by NetProphet 2.0. The use of DTO on the raw expression data (blue bars) increases the size of both the intersection of the ZEV15 and TFKO (left bar grouping) and their union (right bar grouping). Postprocessing with NetProphet 2.0 (green bars) further increases the number of acceptable TFs.