Biomarker Description
TGx-HDACi and TGx-DDI Biomarker Descriptions
The toxicogenomics (TGx) DNA Damage Induction (DDI) biomarker and development is described in the TGx-DDI Classification tool.
The TGx Histone Deacetylase inhibition (HDACi) biomarker consists of 81 signature genes that identify chemicals with or without the potential to induce HDACi in TK6 cells. The signature was derived using the nearest shrunken centroid (NSC) method and expression profiles from TempO-Seq ( https://www.biospyder.com/ ) human whole transcriptome data from TK6 cells exposed for 4 hours to 20 reference compounds, 10 HDACi and 10 non-HDACi (NHDACi). The signature chemical set was limited to inhibitors of the classical human HDACs (Zn2+ dependent HDACs in classes I, II and IV). The probability that the data for each chemical, at the concentration(s) tested, in a test dataset will match the HDACi or the NHDACi signature is predicted using NSC analysis.Test data were prepared as tab-delimited text files containing log2 ratio expression data, normalized to solvent controls, from TempO-Seq Human Whole Transcriptomes (see `Example Data Files` link under `Classification Tool` option on the menu bar). Each column lists data for the different chemicals and/or concentrations tested for all genes detected by TempO-Seq. In addition to the classification p value, six other data outputs are generated for each column of data. The output data include Heat maps, Dendograms (cluster analysis), Principal Component Analysis (PCA), Fold change data, and Gene Cluster and Chemical Cluster distance data. An online Results table is displayed that has an option to view and download the data (see `Example Data Files` link under `Classification Tool` option on the menu bar).
Classification Process
The development of the TGx-HDACi Biomarker and processes underlying the data analysis are described in Cho et al., 2021 (see Publication page). Briefly, the probability that the test data profile matches the profile of either the HDACi or NHDACi reference compounds is determined. The analysis applies the Nearest Shrunken Centroids (NSC) method with a 90% probability cut-off for class membership and the statistical and bioinformatics tools described by Tibshirani et al., 2002. The classification tool assigns a positive HDACi call when the probability is > 0.9 that the test data match the HDACi reference data. When the probability is > 0.9 that the test data match the NHDACi reference data, a NHDACi call is made. If neither probability is >0.9, the classification call is Inconclusive (INC). Results from PCA and hierarchical clustering are also presented to aid the user in the interpretation of the results. The prcomp function in R is used for PCA analysis and the R function hclust is used for hierarchical clustering by Euclidean distances with average linkage. Principle components are estimated using only the training data set.
- prcomp
- principal components function in R
- hclust
- hierarchical clustering function in R