Deep Learning Model and Grad-CAM Visualization of Cancer Gene Expression Classification and Analysis

This paper aims to explore the proposed solution of using deep learning model in genomic data associated in exploring cancer biomarkers. Our proposed solution is the saliency map and functional analysis of the classification task of the 33 cohorts of cancer types extracted in The Cancer Genome Atlas (TCGA) Genome Data Analysis Centre (GDAC) Firehose. The cancer gene expression data is an analyzed microarray transformed into 2-dimensional (2D) images prepared for the experiment model using a convolution neural network. The classification task of the multiclass cancer type is investigated in the deep learning model and the classification accuracy is evaluated. The classification result is examined through a functional analysis to determine if the genes are significant and interpret the analysis through a heat map pattern for the matching of the cancer biomarkers. The utility of the proposed solution is an interpretable outcome of the deep learning model in cancer classification and using gradient-weighted activation maps towards cancer biomarker through visualization.