Since the 1920s, brain tumors have been classified using histological changes that differentiate tumors from healthy cells. This approach has expanded to include molecular classification systems based on characteristics ranging from specific genomic mutations to predictors of chemotherapy success.

To help integrate these strategies for clinical use, researchers developed a technique to synthesize large and disparate datasets and identify similarities across numerous patients. The authors’ approach relies on visualizing sample similarity using multidimensional scaling, a statistical tool that characterizes correlations in the data as distances in space.

Analyzed in this manner, seemingly unrelated factors cluster together on plots, pointing to potentially clinically significant connections. Targeting genome-wide single nucleotide alterations, copy number alterations, DNA methylation, and RNA expression, the authors applied the technique to a combined dataset of glioblastoma and lower grade glioma genomic maps from the Cancer Genome Atlas, a repository of human cancer genome sequences.

The approach revealed that the traditional histologic approach to classifying gliomas imperfectly predicts the molecular composition of these tumors. According to the authors, sample similarity plots help uncover tumor populations that are enriched in specific therapeutic molecular targets.

http://www.pnas.org/content/early/2016/04/25/1601591113