The Categorization of Kidney Cancer Using Machine Learning Based on lncRNA Expression

Overview

While only 1% of the human genome codes for proteins, the other 99% is not junk. Non -coding RNAs are an important factor in cellular function, controlling the transcription of genes, degradation of mRNAs, and other processes. Due to their many cellular functions, long non-coding RNAs (lncRNAs, >200 nt) influence a multitude of diseases and processes. Particu larly, variation in expression of many lncRNAs has been found to be highly specific to various types of cancer. For certain cancer types, specific lncRNAs are up or down regulated predictably. The aberrant expression gives this category of RNA great potent ial as a biomarker to improve the accuracy of cancer diagnosis. Tools such as machine learning can be used to predict a class based on input data. To apply this idea to cancer diagnostics, I trained a logistic regression model to predict kidney cancer subt ype based on lncRNA expression levels. I performed a Principal Component Analysis on lncRNA data obtained from TANRIC, an open resource RNA-seq database. When used to distinguish different types of kidney cancer, the model achieved an accuracy score of 92%. Machine learning models like this could replace current biopsy methods that can be inaccurate. This successful model provides promise for the use of lncRNAs in clinical diagnostic use once more research and data are collected by the scientific community.