A comparison of Principal Component Analysis (PCA) and t-SNE for identifying differences in the patterns of expression of ~80 ribosomal protein transcripts. Each point represents the integrated expression patterns of a single tumor from the indicated set of neoplasms. Note that, because t-SNE is much better able to evaluate non-linear relationships that are PCA, it serves as a better way to express the differences in patterns. t-SNE was able to clearly identify 30 different categories of tumors and distinguish them from their corresponding normal tissues with >95% accuracy.
Ribosomes, the cellular organelles responsible for translating mRNAs into proteins, are comprised of four rRNAs and~80 ribosomal proteins (RPs). Although canonically assumed to be maintained in stoichiometrically equivalent proportions, some RPs have been shown to possess differential expression across tissue types. Dysregulation of RP expression occurs in a variety of human diseases, notably in many cancers where altered expression of single RPs sometimes correlate with different molecular phenotypes and patient survival. Prior studies analyzing RP expression in human cancers have focused on individual RPs without accounting for more complex patterns of variation. Recently, we have utilized a relatively new algorithm for dimensionality reduction known as t-distributed stochastic neighbor embedding (t-SNE) to analyze the patterns of RP transcript (RPT) expression in nearly 10,000 tumors spanning 30 human cancer types along with >700 corresponding normal tissues. We showed that normal tissues and cancers possess readily discernible patterns of RPT expression (see figure below). In tumors, this patterning is distinct from that observed in normal tissues, distinguishes numerous tumor subtypes from one another, and correlates with a variety of molecular, pathological, and clinical features, including survival. Collectively, our results define heretofore unappreciated patterns of RPT expression that predictably occur in normal and malignant tissues. They further demonstrate a powerful and novel method of tumor classification and offer a potential clinical tool for prognosis and therapeutic stratification. Current work is focused on whether other families of functionally related transcripts provide similar predictive power and whether their patterns of expression can be combined with those encoding RPs to provide even more sensitive and specific diagnostic and predictive tools. We are also examining whether perturbing the levels of expression of individual RPs affects tumorigenesis in the mouse model of hepatoblastoma described in Project 1.