Clustering Digestive Tract Tumors Using Transcriptomic and Mutation Data

Digestive tract cancers, like most other cancers, are usually categorized based on cell or tissue of origin. Molecular clustering based on the transcriptome often produces the same classification.

We developed a new method, Newmanization, to reduce underlying tissue signals from transcriptomic analysis. To test our method, we downloaded data on 1635 samples of digestive tract cancers from The Cancer Genome Atlas. The available data includes transcriptomic data by RNA-Seq, as well as binary mutation allele frequency data by whole exome sequencing. We compared, using silhouette widths and visualization by dimension reduction plots, the effectiveness of Newmanized transcriptome and mutation data to separate digestive tract cancers.

The Newmanized transcriptome clusters have clearer separation and larger average silhouette widths. Feature analysis of each cluster for Newmanized transcriptomic data and mutation data revealed that clusters determined with Newmanized data contained more mRNAs present at higher frequencies than clusters defined by mutation data.

This suggests that the Newmanized method holds great potential for advancing personalized transcriptomic medicine.