(A) Visualization of MINDy output. Gene expression profiles were displayed with genes on rows and samples on columns. Expression values for each gene are rank transformed, median centered and rescaled between [−1, 1]. Samples were partitioned based on the expression levels of STK38, and within each partition, sorted by the expression levels of MYC.
(B) ST486 cells were infected with lentivirus expressing shRNA specific to STK38 and collected 96 hours post-infection. Whole cell extracts were analyzed by Western blotting using anti-MYC, anti-STK38/NDR1, and anti-β-Actin antibodies. Representative results from two of five independent experiments are shown.
(C) qRT-PCR analysis of MYC and MYC target genes expression after silencing of STK38 in ST486 cells. Relative expression fold change between STK38 shRNA and control shRNA transduced cells was normalized to that of β-Actin housekeeping gene. Bars represent mean ± standard error (SE) of three different samples.
(D) STK38 mediates MYC phosphorylation. Top graph shows accumulation of phosphorylated MYC in STK38 silenced cells in the presence of proteasomal degradation inhibitor MG132 (48 hours post-infection). ST486 cells were infected with lentivirus expressing STK38 shRNA or control shRNA. The cells were treated with or without MG132 for 4 hours before harvesting. Whole cell extracts were analyzed by Western blotting using anti-STK38, anti-phospho-MYC (Thr58/Ser62), anti-MYC and anti-GAPDH antibodies. Bottom plot provides detailed densitometry histogram. Relative ph-MYC expressions (i.e. ratio of ph-MYC to total MYC) are normalized to that of cells transduced with control shRNA.
(E) HA-STK38 expression vector was transiently transfected into 293T cells with FLAG-MYC expression vector. At 24-hour post-transfection immunoprecipitation was performed using the anti-HA agarose beads. Whole cell extracts and immunoprecipitated proteins were analyzed by Western blotting with anti-FLAG and anti-HA antibodies.
(F) Nuclear extracts from Ramos cells were immunoprecipitated with anti-MYC antibody or mouse IgG as a control. The precipitates were analyzed by Western blotting with anti-STK38/NDR1, anti-MYC, and anti-MAX antibodies.

(A) Visualization of MINDy output. See Figure 2A for interpretation of this graph.
(B) TERT promoter-driven luciferase reporter construct, TERT-800Luc, was transiently co-transfected into 293T cells with MYC plasmid and three different doses of BHLHB2-FLAG plasmids. pRL-TK plasmid was used as internal control to monitor the transfection efficiency. The luciferase activities were measured 48 hours post transfection and normalized against the Renilla activity. Each experiment was done in duplicate and data were shown as mean ± SE of three independent experiments.
(C) qChIP assays on ODH-III cells were performed in parallel using equivalent number of cells. Chromatin was immunoprecipitated with anti-MYC and anti-BHLHB2/DEC1 antibodies or an irrelevant antibody (rabbit IgG) as a control. The precipitated DNA fragments were assessed by quantitative PCR and shown as mean ± SE.
(D) GSEA enrichment analysis for BHLHB2 silencing. Gene expression profiles were generated for 5 samples of ODH-III cells transduced with BHLHB2-specific or control shRNA using a lentiviral vector. The x-axis represents all probes on the microarray, ranked by their absolute differential expression in the cell transduced with the BHLHB2 shRNA vs. the control shRNA. Most differentially expressed genes are towards the left. Sorting was based on the value of −log₁₀ of their differential expression p-values. Blue vertical bars represent all 340 genes in the MYC target signature, while the red bars represent the 80 MINDy-inferred, BHLHB2-modulated subset signature. Color intensity is proportional to the density of the bars. Rank of BHLHB2 among differentially expressed genes is shown by a tick mark. (See Methods for details on GSEA test and statistical significance assessment.)

(A) Visualization of MINDy output. See Figure 2A for interpretation of this graph.
(B) HA-MEF2B expression vector was transiently transfected into 293T cells with FLAG-MYC expression vector. At 48-hour post-transfection, immunoprecipitation was performed using the anti-FLAG agarose beads (M2). Whole cell extracts and immunoprecipitation eluates were analyzed by Western blotting with anti-FLAG, anti-HA, and anti-MAX antibodies.
(C) Nuclear extracts from P3HR1 cells were immunoprecipitated with anti-MEF2B serum or rabbit serum as a control. The precipitates were analyzed by Western blotting with anti-MEF2B and anti-MYC antibodies.
(D) TERT promoter-driven luciferase reporter construct, TERT-800Luc, was transiently co-transfected into 293T cells with MYC plasmid and three different amounts of HA-MEF2B plasmids. pRL-TK plasmid was used as internal control to monitor the transfection efficiency. The luciferase activities were measured 48 hours post-transfection and normalized against the Renilla activity. Each experiment was done in duplicate and data were shown as mean ± SE of three independent experiments.
(E) GSEA enrichment analysis for MEF2B silencing. Gene expression profiles were generated for 5 samples of P3HR1 cells transduced with MEF2B-specific or control shRNA using a lentiviral vector. See Figure 3D for further interpretation of this graph.

(A) Visualization of MINDy output. See Figure 2A for interpretation of this graph.
(B) FLAG-HDAC1 expression vector was transiently transfected into 293T cells with HA-MYC expression vector. At 48-hour post-transfection, immunoprecipitation was performed using the anti-FLAG agarose beads (M2). Whole cell extracts and immunoprecipitation eluates were analyzed by Western blotting with anti-FLAG and anti-HA antibodies.
(C) Nuclear extracts from Ramos and P3HR1 cells were immunoprecipitated with anti-MYC antibody or rabbit IgG as a control. The precipitates were analyzed by Western blotting with anti-HDAC1, anti-MYC and anti-MAX antibodies.
(D) GSEA enrichment analysis for HDAC1 silencing. Gene expression profiles were generated for 5 samples of P3HR1 cells transduced with HDAC1-specific or control shRNA using a lentiviral vector. See Figure 3D for further interpretation this graph.
(E) FLAG-MYC and CBP-HA expression vectors were transiently transfected into 293T cells to purify acetylated-MYC protein. HDAC1, which was also purified from transient transfected 293T cells, was incubated with acetylated-MYC protein in vitro. The signal intensities of acetylated-MYC and total amount of MYC were measured by the imageQuant 5.2 software. The densitometry histogram on the right shows the percent of acetylated-MYC normalized by total amount of MYC.
(F) qChIP assays on P3HR1 cells were performed in parallel using equivalent number of cells. Chromatin was immunoprecipitated with anti-MYC and anti-HDAC1 antibodies or an irrelevant antibody (rabbit IgG) as a control. The precipitated DNA fragments were assessed by quantitative PCR with CR2 and p21^CIP1 primers and are shown as mean ± SE.

(A) A gene expression profile dataset is represented as a matrix, where columns indicate different samples and rows indicate different genes. Given a transcription factor of interest, TF, a modulator and a target to be tested by MINDy (M and t respectively) are chosen among the remaining genes. Some modulator-target combinations may be eliminated a priori based on functional or statistical constraints (blue rectangle). For instance, one may want to consider only ubiquitin ligases as candidate modulators. All the samples are then sorted according to the expression of the selected modulator M. The set of samples (e.g., 35%) with the lowest and highest expression of the modulator are then selected. These sample sets are labeled M-low and M-high. In each of the two sample sets, samples are finally sorted according to the TF expression. Three cases are possible when comparing the TF-target correlation in M-high vs. M-low. Scenario 1 (Positive Inferred Modulator): Significant increase in MI (i.e. more correlation in the M-high set than in the M-low set); Scenario 2 (Negative Inferred Modulator): Significant decrease in MI (i.e. less correlation in the M-high set than in the M-low set); Scenario 3&4 (Not a Modulator): No significant MI change is observed.
(B) Synthetic network analysis using MINDy. TF: the transcription factor of interest; K: a protein kinase; cTF: a co-transcription factor; : activated TF through phosphorylation by K; : transcriptionally active complex formed between TF and cTF; G₁, G₂ and G₃ are three downstream targets of TF. More details on the synthetic network are provided in the Methods.
(C) Networks reconstructed by ARACNe and MINDy. ARACNe (left) is based on pairwise MI, thus captures only the unconditional interaction between TF and G₁. MINDy, on the other hand, infers three-way interactions using the CMI. When K is used as candidate modulator (middle), it correctly identified the conditional interaction between TF and G₃; similarly, it also correctly inferred cTF as the modulator of TF-G₂ interaction (right). Interactions are labeled with their respective p-values. Interactions that are statistically significant, after correction for multiple testing, are shown in black with red p-values.