We developed software that uses RP expression patterns to accurately diagnose cancer and predict the odds of survival for individual patients.
Description
Perturbations of ribosomal protein (RP) expression are associated with numerous human cancers, including breast, pancreas, bladder, and brain. Past attempts to summarize the heterogeneity of RP expression in human cancers was limited to single RPs among cancer cohorts, without accounting for larger patterns of variation that might distinguish tumors from one another. So far, prognostic tests for molecular markers of metastasis have been limited to a single type of cancer, such as the MammaPrint test for early stage breast cancer. To identify and classify RP transcript patterns, we applied an advanced form of machine learning called T-distributed stochastic neighbor embedding (T-SNE) that uses a variety of linear and non-linear relationships to cluster data. When applied to human tissue data from the cancer genome atlas, this method was 93% accurate at distinguishing between tissue types and more than 98% accurate at discriminating tumors from normal tissue. In at least ten different common tumors typesincluding hepatocellular carcinoma, kidney, brain and endometrial cancer,the pattern of RP transcripts was also highly predictive of survival. Our proprietary T-SNE-based RP transcript analysis program could form a clinically useful bioinformatics platform to accurately determine a tumor’s tissue of origin, classify known tumors into subtypes, and stratify patients into high-andlow-risk categories. This information will be useful for determining the most appropriate treatment plan for individual patients.
Applications
· Guiding treatment decisions for many different types of cancer
· Determining diagnosis as well as prognosis
Advantages
· Widely applicable to many different types of cancer
· More accurate at determining clinically relevant subtypes than current methods
· Highly predictive of survival
Invention Readiness
Software
IP Status
https://patents.google.com/patent/WO2019018374A1
