Gene Expression Based Tumor Purity Estimation and Individual-Specific Survival Tool in Skin Cutaneous Melanoma

Abstract

Skin Cutaneous Melanoma (SKCM) is a type of cancer that arises from the occurrence of genetic mutations in melanocytes and is the most aggressive and fatal type of cancer affecting mostly the Caucasian population with increasing incidences in Asia. Tumors and lesions are highly heterogeneous comprised of cancerous and non-cancerous cells, and the admixture is thought to have an important role in tumor growth and progression of the disease. This study features a system capable of estimating tumor purity from RNA-Seq gene expression data using Gradient Boosting Machines and providing individual-specific survival prediction (death or progression of the disease) using a set of clinical features and the tumor purity estimate from the trained model. The performance of the models for tumor purity using the entire set of gene expression and selected features by importance scores were compared. The survival models have shown that the tumor purity estimate from the trained model provided additional prognostic information over established clinical features including age, tumor stage, and sex. Survival models using Cox Proportional Hazards are provided to allow users to evaluate and probe the models for further in-sights, whether with past historical cases, current or hypothetical patients. Future model improvements and prospective replication will be necessary to demonstrate true clinical utility.