One-Class SVM to Identify Candidates to Reference Genes Based on the Augment of RNA-seq Data with Generative Adversarial Networks

Rueda, Edwin J.; Ramos, Rommel; Franco, Edian F.; Belo, Orlando; Morais, Jefferson

Reference Genes (RG) are constitutive genes required for the maintenance of basic cellular functions. Different high-throughput technologies are used to identify these types of genes, including RNA sequencing (RNA-seq), which allows measuring gene expression levels in a specific tissue or an isolated cell. In this paper, we present a new approach based on Generative Adversarial Network (GAN) and Support Vector Machine (SVM) to identify in-silico candidates for reference genes. The proposed method is divided into two main steps. First, the GAN is used to increase a small number of reference genes found in the public RNA-seq dataset of Escherichia coli. Second, a one-class SVM based on novelty detection is evaluated using some real reference genes and synthetic ones generated by the GAN architecture in the first step. The results show that increasing the dataset using the proposed GAN architecture improves the classifier score by 19%, making the proposed method have a recall score of 85% on the test data. The main contribution of the proposed methodology was to reduce the amount of candidate reference genes to be tested in the laboratory by up to 80%.