2D representation of transcriptomes by t-SNE exposes relatedness between human tissues

Abstract

The GTEx Consortium reported that hierarchical clustering of RNA profiles from 25 unique tissue types among 1641 individuals accurately distinguished the tissue types, but a multidimensional scaling failed to generate a 2D projection of the data that separates tissuesubtypes. In this study we show that a projection by t-Distributed Stochastic Neighbor Embedding is inline with the cluster analysis which allows a more detailed examination and visualization of human tissue relationships.