filtrate
(Package: pbcmc) :
Virtual functions for MolecularPermutationClassifier hierarchy
The following functions establish an organized framework for MolecularPermutationClassifier subclasses data processing. In this context, the later are supposed to be implemented with respective responsibilities. In particular, once the class is created the user has to:
This is a concrete MolecularPermutationClassifier based on Perou et al. (2000 & 2010) PAM50 molecular signature, using genefu package implementation (Haibe-Kains et al. 2014).
Obtain PAM50 subtype using genefu centroid Spearman's correlation implementation. If std=="median" probes with the same mapping are averaged. Then, the complete database is center normalized using gene median expression. This is done in order to assure selecting the same "gene" to those in "genefu" library, instead of the most variant probe (default in geneid.map), when more than one probe match the same gene. This selection is based on probe population variance that could depends on the number of accounted genes.
Remove exprs rows not required by MolecularPermutationClassifier subclasses to classify samples, in this case PAM50. This means to only keep genes with valid EntrezGeneID, i. e., not NA and present in PAM50 signature centroids. In addition, annotation slot will only keep "probe", "EntrezGene.ID" and "NCBI.gene.symbol" fields required by genefu's intrinsic.cluster.predict function.
Calculate the null Spearman's ρ distribution of each subtype by means of gene label permutation, in order to evaluate if the observed values could be obtained by random change.