A Scalable Big Data–Enabled Deep Learning Architecture for Integrated Analysis of DNA and RNA Sequencing Data in Cancer Genomics

Abstract

Cancer is a very heterogeneous condition provoked by complicated interplay between the changes on a genetic level and transcriptional regulation, and it is difficult to properly characterise and predict without the involvement of both single-omics data sets. The high dimensionality, nonlinearity and scale of these new sequencing data can be difficult to say the least with traditional statistical and machine learning methods. In order to overcome these shortcomings, this paper suggests a scalable deep learning solution based on big data analysis to combine DNA and RNA sequencing to analyse cancer genomics. In the given framework, a multi-omics fusion strategy, i.e., the combination of DNA- and RNA-based features, is learned via a deep neural network (DNN): in this framework, a set of encoding branches is used to learn features based on DNA and RNA sequences, and a learned combination is created to create a unified predictor. The effectiveness of the proposed approach is tested using publicly available cancer genomics data, such as those of The Cancer Genome Atlas (TCGA). The experimental evidence shows that the integrated DNA RNA model is always better than the single-omics models, as it provides a better accuracy, F1-score, and area under the ROC curve (AUC) in various evaluation environments. These results demonstrate the usefulness of deep learning-based multi-omics integration to provide complementary information on the molecular level and best predictive outcomes. The suggested architecture in general offers a highly scalable, and extended architecture on integrative cancer genomics analysis, which may find potential applications in translational oncology, disease stratification, and precision clinical decision making, supported by data.