Skip to main content
. 2018 May 16;9:176. doi: 10.3389/fgene.2018.00176

Table 4.

Step-by-step output as numbers of qualifying genes along the LSTNR analytical pipeline for liver transcriptomes from male Sprague-Dawley rats after toxicant exposure based on the mode-of-action training RNAseq dataset by the MAQC phase III SEQC crowdsource toxicogenomics (TGxSEQC) effort (GEO accession number: GSE55347).

Criteria Hepatotoxicity: Mode-of-Action Rat Models (N = 54)
Genes with uniquely aligned reads 30,852
Distribution of gene-wise RPM means P(y)~Weibull3P(y;α,ß,γ);
y = RPM
α = 6.7 RPM
ß = 0.38
γ = 2.5 × 10−3 RPM
Independent filtering: Genes with average RPM y > α 9,593
Linearized normalizing transformant: GLM Linear Predictor (y–γ)−1
Transformant two-way ANOVA: resolved genes across groups with respect to gene-wise mean 3,975
Resolution-Weighed ANOVA: Significant Genes (SGs) with FDR adj. p < 0.05 based on differences in resolution-weighed RPM log-fold changes (Log2FC) relative to baseline condition 5,983
Differential expression: DEGs = subset SGs that exhibit both:

  • resolution-weighed effect size above 5% of gene-wise variation (δLog2FC>0.3 × σSSR); and

  • post-hoc pairwise-significant Log2FC differences between at least two MOAs (Student's t-test p < 0.05)

 5,864
Reproducibility: LSTNR genes = subset of SGs that exhibit both:

  • resolution-weighed effect size above 5% of gene-wise variation (δLog2FC>0.3 × σSSR); and

  • at least one group with Log2FC differences vs. baseline greater than 95% Tolerance Interval of gene × group residuals among SGs (post-hoc pairwise-significance not required)

 1,953
Expectable DEGs: DEGREEs = Ensembl-annotated DEGs with a reproducible expectation estimate (i.e., DEGs that are also LSTNRs) and official Entrez symbol 1,510
Transcriptional profiling: Profiler DEGREEs = top DEGREEs ranked by retrospective statistical power with monotonically decreasing within-gene effect sizes ΔLog2FC 65 Profiler DEGREEs
Diagnostic targets: Biomarkers = minimal subset of Profiler DEGREEs with predictive discriminant power based on sequential partition tree analysis (ROC scores>0.9 per phenotype) Ucp3, Tmem86b, Sugct, Acaa1b, Hadhb, Tfam, Acaa1a, and Gsdmd