This function fits the Subtype Clustering Model as published in Desmedt et al. 2008 and Wiarapati et al. 2008. This model is actually a mixture of three Gaussians with equal shape, volume and variance (see EEI model in Mclust). This model is adapted to breast cancer and uses ESR1, ERBB2 and AURKA dimensions to identify the molecular subtypes, i.e. ER-/HER2-, HER2+ and ER+/HER2- (Low and High Prolif).
Matrix containing the ESR1-related gene(s) in rows and at least three columns: "probe", "EntrezGene.ID" and "coefficient" standing for the name of the probe, the NCBI Entrez Gene id and the coefficient giving the direction and the strength of the association of each gene in the gene list.
module.ERBB2
Idem for ERBB2.
module.AURKA
Idem for AURKA.
data
Matrix of gene expressions with samples in rows and probes in columns, dimnames being properly defined.
annot
Matrix of annotations with at least one column named "EntrezGene.ID", dimnames being properly defined.
do.mapping
TRUE if the mapping through Entrez Gene ids must be performed (in case of ambiguities, the most variant probe is kept for each gene), FALSE otherwise.
mapping
**DEPRECATED** Matrix with columns "EntrezGene.ID" and "probe" used to force the mapping such that the probes are not selected based on their variance.
do.scale
TRUE if the ESR1, ERBB2 and AURKA (module) scores must be rescaled (see rescale), FALSE otherwise.
rescale.q
Proportion of expected outliers for rescaling the gene expressions.
do.BIC
TRUE if the Bayesian Information Criterion must be computed for number of clusters ranging from 1 to 10, FALSE otherwise.
model.name
Name of the model used to fit the mixture of Gaussians with the Mclust from the mclust package; default is "EEI" for fitting a mixture of Gaussians with diagonal variance, equal volume, equal shape and identical orientation.
plot
TRUE if the patients and their corresponding subtypes must be plotted, FALSE otherwise.
filen
Name of the csv file where the subtype clustering model must be stored.
verbose
TRUE to print informative messages, FALSE otherwise.
Value
model
Subtype Clustering Model (mixture of three Gaussians), like scmgene.robust, scmod1.robust and scmod2.robust when this function is used on expO dataset (International Genomics Consortium) with the gene modules published in the two references cited below.
BIC
Bayesian Information Criterion for the Subtype Clustering Model with number of clusters ranging from 1 to 10.
subtype
Subtypes identified by the Subtype Clustering Model. Subtypes can be either "ER-/HER2-", "HER2+" or "ER+/HER2-".
subtype.proba
Probabilities to belong to each subtype estimated by the Subtype Clustering Model.
subtype2
Subtypes identified by the Subtype Clustering Model using AURKA to discriminate low and high proliferative tumors. Subtypes can be either "ER-/HER2-", "HER2+", "ER+/HER2- High Prolif" or "ER+/HER2- Low Prolif".
subtype.proba2
Probabilities to belong to each subtype (including discrimination between lowly and highly proliferative ER+/HER2- tumors, see subtype2) estimated by the Subtype Clustering Model.
module.scores
Matrix containing ESR1, ERBB2 and AURKA module scores.
Author(s)
Benjamin Haibe-Kains
References
Desmedt C, Haibe-Kains B, Wirapati P, Buyse M, Larsimont D, Bontempi G, Delorenzi M, Piccart M, and Sotiriou C (2008) "Biological processes associated with breast cancer clinical outcome depend on the molecular subtypes", Clinical Cancer Research, 14(16):5158–5165.
Wirapati P, Sotiriou C, Kunkel S, Farmer P, Pradervand S, Haibe-Kains B, Desmedt C, Ignatiadis M, Sengstag T, Schutz F, Goldstein DR, Piccart MJ and Delorenzi M (2008) "Meta-analysis of Gene-Expression Profiles in Breast Cancer: Toward a Unified Understanding of Breast Cancer Sub-typing and Prognosis Signatures", Breast Cancer Research, 10(4):R65.