Using Gene-level to Generalize Transcript-level Classification Performance on Multiple Colorectal Cancer Microarray Studies

Abstract

Several classification algorithms have been applied into microarray studies for colorectal cancer identification. Algorithms such as naive bayes, random forest, logistic regression, support vector machine, and deep learning have been successfully used in previous studies. The accuracy of these algorithms shown promising result through n-fold validation. However, most of studies are limited to transcript-level that will implicate to biased interpretation of classification result due to different microarray platform entanglement. Therefore, we applied gene-level classification to generalize transcript-level classification result on multiple colorectal cancer microarray studies through different classification algorithms including: naive Bayes, random forest, logistic regression, support vector machine, and deep learning. We evaluated classification performance using several parameters including: accuracy, area under ROC curve, recall and precision. As the result, we found biased classification result in transcript-level from multiple microarray studies can be solved through gene-level classification by applying annotation and merging. In addition, applying batch effect removal method can make gene-level classification performance slightly improved. Furthermore, annotation and merging also can be used to solve another biased result of feature selection in transcript-level.