Table 1.
Table 2.
DrugMatrix [21], [22] structure-activity classified drugs and toxicants.
Figure 1.
Iterative signature algorithm (ISA) module generation analysis.
Number of iterative signature algorithm (ISA) modules NM as a function of the number of starter gene sets, Nstarter.
Figure 2.
Module specificity and intra-module correlation.
A) Tests for maximum module specificity, and B) maximum intra-module gene correlation. ISA, iterative signature algorithm gene co-expression modules; HC, hierarchical clustering gene sets; MAZ, gene sets composed of the top differentially expressed genes associated with injury indicator; SVM, support vector machines gene sets; PPI, protein-protein interaction network gene sets; RAND, gene sets composed of 100 genes selected at random. All method results were statistically significantly (p-value <0.05) different from the results generated using the random gene set.
Figure 3.
Clustering and analysis of injury indicators using module activation patterns.
A) Correlation among injury indicators. The clinical endpoints used in Table 1 are not independent, but highly correlated both from a biological point of view as well as from the gene transcription activation pattern. The hierarchical clustering dendrogram identifies the most related endpoints based on a Pearson correlation of iterative signature algorithm (ISA) module activation patterns. B) Correlation of injury indicators with structure activity classes. The clustering of the indicators is shown by a dendrogram at left; at center are the various injury indicators; at right is a heat map with elements equal to the Pearson correlation between the injury indicators at center, and the structure activity indicators arrayed across the top right. The Pearson correlation is determined using the covariance of the ISA module activation patterns of the injury indicators and structure activity classes. The Pearson correlation value in the first column of the table is the average intra-cluster correlation between indicators in the same cluster. CP, clinical pathology; LH, liver histopathology; OW, organ weight.
Figure 4.
Clustering of the iterative signature algorithm (ISA) modules.
By construction, the modules represent groups of genes co-expressed across a subset of the conditions, and they may share genes. The clustering gauges the independence of the modules and groups some modules together. A dendrogram of the clustering is shown at right, giving the module membership 1–78 of each of the 28 module clusters. The Pearson correlation is the average intra-cluster correlation between modules in the same cluster.
Figure 5.
Activation pattern of module clusters.
5A) Reduced representation of the each module cluster's activation patterns for the injury indicators shown in Figures 3 and 4. The illustration highlights the differences and similarities of each injury indicator based on their module activation patterns. 5B) Shows the root-mean-square distance between all unique injury-indicator cluster pairs calculated using the averaged activation scores .
Table 3.
KEGG pathway mapping.
Figure 6.
Activation patterns for selected modules and biomarker genes.
Activation patterns shown correspond to the 25 injury indicators in Table 1. Labeled peaks represent average module activation score greater than 1.5 as calculated using Equation (8). A) The top graph shows averaged activation of modules 43 and 44 compared with the gene activation pattern of alanine aminotransferase (Gpt). B) The middle graph shows the average activation of modules 47 to 51 compared to the gene activation pattern of aspartate aminotransferase (Got1). CP, clinical pathology; OW, organ weight.
Figure 7.
Module 55 activation across the 25 injury indicators.
Activations shown represent the 25 injury indicators in Table 1. Labeled peaks represent a module 55 activation score greater than 1.5 as calculated using Equation (8). CP, clinical pathology; LH, liver histopathology.
Figure 8.
Activation of selected genes from Module 55.
Selected genes show significant gene activation for fibrotic conditions. Lcn2, lipocalin 2; Lbp, lipopolysaccharide binding protein; A2m, alpha-2-macroglobulin; Ltb, lymphotoxin beta; Pcolce, procollagen C-endopeptidase.
Figure 9.
Module activation patterns for periportal lipid accumulation and periportal fibrosis.
Module activation patterns for A) Periportal lipid accumulation and B) Periportal fibrosis. The grey box represents an absolute module activation score greater than 1.5 as calculated using Equation (8). Activation scores greater than the cut-off are labeled by their associated module numbers and module clusters. Modules are labeled with their center genes if the genes have a curated association with liver injury in the Comparative Toxicogenomics Database (*), if the genes code for secreted proteins (†), or if the genes are shared between periportal lipid accumulation and periportal fibrosis (‡). Modules are also labeled with member genes (not necessarily center genes) if they are part of the FibroSure biomarker set (#).
Table 4.
Gene signatures for Periportal lipid accumulation.
Table 5.
Gene signatures for Periportal fibrosis.
Table 6.
Selected general liver injury signature genes with known disease annotations in the Comparative Toxicogenomics Database [20].
Figure 10.
Validation of external datasets.
Scatter plots show the correlation of the log-ratios between DrugMatrix data and external datasets for the periportal fibrosis gene signature. Comparison of the log-ratios in DrugMatrix periportal fibrosis conditions with A) 15 mg/kg of naphthyl isothiocyanate at four days of exposure obtained from the Toxicogenomics Project-Genome Assisted Toxicity Evaluation System (TG-GATEs), B) 15 mg/kg of naphthyl isothiocyanate at eight days of exposure obtained from TG-GATEs, C) 15 mg/kg of naphthyl isothiocyanate at 15 days of exposure obtained from TG-GATEs, and D) liver fibrosis produced by bile duct ligation obtained from GSE13747.
Figure 11.
Analysis of exposures in GSE5509 using the general liver injury gene signature.
Multidimensional scaling (MDS) plot of six chemical exposures in GSE5509 using the general liver injury gene signature. This figure shows the ability of the genes in the general liver injury signature to separate toxicants from non-toxicants. Rosiglitazone, caerulin, and di-nitro phenol, the non-toxic compounds in this set are marked in green circles. α-Naphthyl-isothiocyanate, dimethyl nitrosamine, and N-methyl formamide are the toxic compounds in this set, and they are marked with red triangles. In the MDS plot, the non-toxic compounds clustered separately form the toxic compounds. We have highlighted the non-toxic compounds within a green circle.