Deep Learning Based Tumor Type Classification Using Gene Expression Data

Abstract

Differential analysis occupies the most significant portion of the standard practices of RNA-Seq analysis. However, the conventional method is matching the tumor samples to the normal samples, which are both from the same tumor type. The output using such method would fail in differentiating tumor types because it lacks the knowledge from other tumor types. Pan-Cancer Atlas provides us with abundant information on 33 prevalent tumor types which could be used as prior knowledge to generate tumor-specific biomarkers. In this paper, we embedded the high dimensional RNA-Seq data into 2-D images and used a convolutional neural network to make classification of the 33 tumor types. The final accuracy we got was 95.59%, higher than another paper applying GA/KNN method on the same dataset. Based on the idea of Guided Grad Cam, as to each class, we generated significance heat-map for all the genes. By doing functional analysis on the genes with high intensities in the heat-maps, we validated that these top genes are related to tumorspecific pathways, and some of them have already been used as biomarkers, which proved the effectiveness of our method. As far as we know, we are the first to apply convolutional neural network on Pan-Cancer Atlas for classification, and we are also the first to match the significance of classification with the importance of genes. Our experiment results show that our method has a good performance and could also apply in other genomics data.

Transfer learning with convolutional neural networks for cancer survival prediction using gene-expression data

“…This leads us to think that if we rearranged gene-expression data (in an element-wise manner), thus transforming gene-expression samples into better structured patterns-giving them an"image form" by putting together those genes that share domain-specific information-, CNNs could extract local high-level features and achieve higher performance rates when tackling predictive or classification problems. Following this idea, in [37] Lyu and Haque presented the first preliminary approach to transform gene-expression vectors into two-dimensional images, which were subsequently used to train a CNN architecture for tumor-type classification. They utilized the relative position of the genes in the chromosome, i.e.…”

“…Only a few recent studies [37,38] have already explored the idea of using biological criteria to transform gene-expression vectors into structured images. In these preliminary works, gene-expression images were generated in a single step, by directly mapping gene-expression values to a fixed set of colors, using domain-specific information to determine the position of every gene inside the images.…”

Transcriptomics in Toxicogenomics, Part III: Data Modelling for Risk Assessment

“…Algorithms for DL that have been used with success [155] include feedforward neural network (FNN) [156], convolutional neural network (CNN) [157], and graph convolutional network (GCN) [158]. For example, Wang et al [159] compared the performances of deep neural networks (DNN) with respect to RF and SVM in the prediction of chemically induced liver injuries.…”

Deep Learning in Omics Data Analysis and Precision Medicine

“…In addition, this method identified a set of highly interactive genes which could be good cancer biomarkers. Gene expression data from TCGA have also been exploited to accurately differentiate samples into different cancer types (60). On the other hand, Si et al used an AE to classify healthy and breast cancer patients using methylation data (61), while Chatterjee et al used CNN to classify different cancer types by their methylation patterns, achieving very promising results (62).…”