Format

Send to

Choose Destination
J Mol Diagn. 2017 Nov;19(6):881-891. doi: 10.1016/j.jmoldx.2017.07.008. Epub 2017 Sep 1.

Molecular Signatures for Tumor Classification: An Analysis of The Cancer Genome Atlas Data.

Author information

1
Princess Margaret Cancer Centre and MacFeeters-Hamilton Centre for Neuro-Oncology Research, Toronto, Ontario, Canada.
2
Department of Computer Science, University of Toronto, Toronto, Ontario, Canada.
3
Princess Margaret Cancer Centre and MacFeeters-Hamilton Centre for Neuro-Oncology Research, Toronto, Ontario, Canada; Department of Neurosurgery, Toronto Western Hospital, Toronto, Ontario, Canada. Electronic address: gelareh.zadeh@uhn.ca.
4
Princess Margaret Cancer Centre and MacFeeters-Hamilton Centre for Neuro-Oncology Research, Toronto, Ontario, Canada; Department of Laboratory Medicine and Pathobiology, University of Toronto, Toronto, Ontario, Canada. Electronic address: ken.aldape@uhn.ca.

Abstract

Cancer classification in the clinic is primarily based on histological analysis in the proper clinical context, often supplemented by immunohistochemical and molecular studies. Recent genomic studies have shown the potential of integrated multiomics platforms for molecular classification. We performed unsupervised analyses of molecular platforms in The Cancer Genome Atlas data (n = 6,216 samples) in comparison with tumor type. Our data showed that mRNA signatures and DNA methylation signatures mapped to histological diagnosis with high accuracy (95% and 88%, respectively) as individual platforms. The accuracy of mRNA signatures alone for classification and subtype identification was comparable to accuracies reported in the previously published Pan-Cancer 12 analysis. When combined, mRNA and methylation revealed a set of outliers for which the mRNA- and methylation-based molecular signatures concordantly differed from the original histological diagnosis. A subset remained consistent as outliers after using alternative classification and clustering algorithms and analysis of an independent molecular platform (miRNA). Overall, our results indicate that unsupervised approaches with individual genomic platforms, especially gene expression and DNA methylation, provide substantial classification information and identify occasional outlier cases in which the molecular signature is distinct from signatures expected for a given histological diagnosis. Identification of cases in which the molecular signature correlates with a specific histology that differs from initial impressions may prompt reconsideration of tumor classification in specific cases.

PMID:
28867603
DOI:
10.1016/j.jmoldx.2017.07.008
[Indexed for MEDLINE]

Supplemental Content

Full text links

Icon for Elsevier Science
Loading ...
Support Center