Convolutional neural network models for cancer type prediction based on gene expression

Abstract

BackgroundPrecise prediction of cancer types is vital for cancer diagnosis and therapy. Important cancer marker genes can be inferred through predictive model. Several studies have attempted to build machine learning models for this task however none has taken into consideration the effects of tissue of origin that can potentially bias the identification of cancer markers. ResultsIn this paper, we introduced several Convolutional Neural Network (CNN) models that take unstructured gene expression inputs to classify tumor and non-tumor samples into their designated cancer types or as normal. Based on different designs of gene embeddings and convolution schemes, we implemented three CNN models: 1D-CNN, 2D-Vanilla-CNN, and 2D-Hybrid-CNN. The models were trained and tested on combined 10,340 samples of 33 cancer types and 731 matched normal tissues of The Cancer Genome Atlas (TCGA). Our models achieved excellent prediction accuracies (93.9-95.0%) among 34 classes (33 cancers and normal). Furthermore, we interpreted one of the models, known as 1D-CNN model, with a guided saliency technique and identified a total of 2,090 cancer markers (108 per class). The concordance of differential expression of these markers between the cancer type they represent and others is confirmed. In breast cancer, for instance, our model identified well-known markers, such as GATA3 and ESR1. Finally, we extended the 1D-CNN model for prediction of breast cancer subtypes and achieved an average accuracy of 88.42% among 5 subtypes. The codes can be found at -3 -https://github.com/chenlabgccri/CancerTypePrediction. ConclusionsHere we present novel CNN designs for accurate and simultaneous cancer/normal and cancer types prediction based on gene expression profiles, and unique model interpretation scheme to elucidate biologically relevance of cancer marker genes after eliminating the effects of tissue-of-origin. The proposed model had light hyperparameters to be trained and thus can be easily adapt to facilitate cancer diagnosis in the future. BackgroundCancer is the second leading cause of death worldwide, an average of one in six deaths is due to cancer [1]. Considerable research efforts have been devoted to cancer diagnosis and treatment techniques to lessen its impact on human health. Cancer prediction's major focus is on cancer susceptibility, recurrence, and prognosis, while the aim of cancer detection is the classification of tumor types and identification of markers for each cancer such that we can build a learning machine to identify specific metastatic tumor type or detect cancer at their earlier stage. With the increased awareness of precision medicine and early detection techniques matured over years of technology development [2][3][4], including particularly many detection screens achieving a sensitivity around 70-80% [5], the demand for applying novel machine learning methods to discover new biomarkers has become one of the key driving factors in many clinical and translational applications.Deep learning (DL), a branch of Artificial ...

Integrated Multi-omics Analysis Using Variational Autoencoders: Application to Pan-cancer Classification

“…The average accuracy of this model was 95.59%. Mostavi et al [18] implemented three CNN models on the same datasets and achieved 95.7% accuracy for 33 tumour types and 95.0% accuracy for tumour types and normal samples. All of these studies only used gene expression data for pan-cancer classification without including complementary information from other types of omics data.…”

“…The average classification accuracy on different tumour types and normal samples (34 classes) achieved by the end-toend OmiVAE is 97.49% after 10-fold cross-validation and the standard deviation is 0.45%. As for the performance of other methods, only Mostavi et al [18] evaluated their model on the 34-class task for both tumour and normal samples with an accuracy of 95.0%. To further visualise the classification space learned by OmiVAE, we used t-SNE to reduce the dimension of latent vectors from 128 to 2 and plotted samples on a 2D scatter graph shown in Fig.…”

OncoNetExplainer: Explainable Predictions of Cancer Types Based on Gene Expression Data

“…Lyu et al [6] and Mostavi et al [17] embedded the RNA-Seq data from the PCA project into 2D images and trained a CNN to classify 33 tumor types, which outperforms the approach in [5]. Besides, they provide a functional analysis on the genes with high intensities in the HM based on GradCAM and validated that these top genes are related to tumor-specific pathways.…”

Machine Learning and Deep Learning challenges for building 2′O site prediction

“…Deep Learning (DL) models have shown unprecedented success in a wide variety of applications ranging from classifying human brain signals to predicting sites of epitranscripome modifications [2,3]. In particular, Convolutional Neural Networks (CNNs) can be extremely beneficial in biological datasets not only by achieving high accuracy in predictive tasks but also by providing a clear explanation of the inner working process of the model [4,5]. Furthermore, automatic feature extraction in DL models is eliminating conventional human-made feature extraction methods that require a well-established prior knowledge about the biological problem of interest.…”