Figures in manuscript
| Figure | Legend | Download |
|---|---|---|
| Figure 1 | NACEP method. NACEP starts with a Dirichlet Process-driven clustering of time-course data. Instead of assigning each gene into a particular cluster, NACEP retains the probabilities of this gene to belong to every cluster. These probabilities and the mean expression patterns of every cluster are used in the next step of comparing the temporal expression patterns of a gene. | (EPS file) (JPEG file) |
| Figure 2 | The Gibbs Sampler algorithm. The details of the updating strategies and the forms of conditional probabilities are provided in the Supplementary Text. | (EPS file) (JPEG file) |
Figure 3 |
Clustering performance. (A) One hundred datasets were simulated. Each dataset contained four clusters. Representative expression values of three genes for each cluster are shown, in +, o and ∆. (B) Clustering performance was evaluted. CNPE: cluster number prediction error, the proportion of predictions with incorrect cluster numbers, among all simulated datasets. A higher CNPE correlates with worse performance. AMR: average misclassification rate, the average proportion of genes being misclassified. A higher AMR correlates with worse performance. | (EPS file) (JPEG file) |
| Figure 4 | Cross-condition comparison. (A) Four synthetic datasets. Five to seven groups of genes were simulated in each synthetic dataset, with each group exhibiting either a different pattern (Groups a-e), or a similar pattern (Groups f, g) between the two experimental conditions. Genes in Groups e and g do not form any clusters. Each gene was generated from its own temporal pattern. Group d differs in the two conditions by a phase shift. The cluster averages and their confidence intervals are shown in solid curves and shaded regions, respectively. Expression values of ten representative genes are shown for each gene group in dots. The standard deviation of the expression values of all genes in a group is shown as a green band. (B) ROC curves. Panels 1-4 correspond to synthetic datasets 1-4. The ROC curves of the best, median, and worst performance on 50 simulations are plotted in solid (NACEP) and dashed (EDGE) lines. | (EPS file) (JPEG file) |
| Figure 5 | Comparison of RA-induced and spontaneous differentiation of ES cells. (A) Mean expression patterns of 37 clusters in days 0-7 of two differentiation conditions. (B) A hypothetical regulatory pathway that responds to RA and induces neural differentiation of ES cells. | (EPS file) (JPEG file) |
| Figure 6 | Comparison of time-course data of knockdown of seven transcription factors. (A) A hypothetical gene regulatory pathway. (B) The top 30 NACEP predicted genes with differential temporal patterns between an RNAi condition and the control. (C) Pearson correlation between TF knockdowns. The Pearson correlation was derived from the NACEP distances (di) of all genes between two TF knockdowns. (D) Predicated relative TF placement in a GRN, drawn with Cytoscape (Shannon, et al., 2003). The pairwise TF correlations are visualized as the thickness of the edges. Dashed edges represent experimentally verified regulatory interactions. | (EPS file) (JPEG file) |
