Bootstrapped ROC (bROC)
bROC algorithm is used in the discovery of differentially expressed probes/genes in microarray and RNA-seq experiments.
bROC plugin deploys in CLC Main Workbench and CLC Genomics Workbench.
ROC (receiver operating characteristic) is a generally applicable, non-parametric procedure that provides insight into the discriminatory properties of data features for a binary classifier. However, the method is not efficient for gene expression experiments as they generally do not produce a sufficient number of samples. bROC overcomes that limitation by resampling (bootstrapping) the expression data to produce a large number of simulated measurements that preserve the statistical properties of the original data.
Thus, bROC can produce detailed curves of sensitivity (probability of true positive detection) vs. 1-specificity (probability of false positive detection) for all features of interest. CONF = 2 AUC - 1, where AUC is the area under ROC curve, is the primary statistics used for detection of regulated features (probes/genes).
Version 3 (August 2013) includes data normalization and graphical outputs.
- Works particularly well for experiments with a small number of experimental/biological replicates.
- Includes data normalization (for RNA-seq, mainly).
- When combined with RNA-seq Analysis (CLC bio), provides complete differential expression analysis workflow for RNA-seq data.
- Graphical outputs facilitate interpretation of results.
- With no user-selectable parameters, the algorithm is easy to use.
- Non-parametric approach is applicable to all platforms producing expression data for a large number of features (transcripts, genes).