Figure 5 : Performance of random forest models for paclitaxel sensitivity prediction in breast cancer patients.

From: In silico Pathway Activation Network Decomposition Analysis (iPANDA) as a method for biomarker development

Figure 5

Models were built in respect to three distinct end points: sensitivity of ERN HER2P cancer type, ERN HER2N type and all breast cancer types mixed. The models were trained using iPANDA, SPIA, GSEA, ssGSEA, PLAGE and DART pathway activation (enrichment) scores and gene-level data including: gene expression values for all genes (logGE), fold changes of tumour samples relative to the mean of paired normal samples for all genes (logFC), gene expression for differential genes only (genes are considered differentially expressed if t-test P value was <0.05 between tumour and normal samples, logDGE) and fold changes for differential genes only (logDFC). MCC, specificity and sensitivity performance metrics are shown for each model.