An overview of biological data generation using generative adversarial networks

Due to the high cost of biological data access and the privacy issues, collecting a large amount of biological data for training deep learning model is difficult in the field of biology. Concerning this issue, this article focuses on generative adversarial networks (GANs), which is a special type of deep learning model, and reviews their representative applications for generating biological data. We briefly introduced the working principle of GAN, and numerous applications to the areas of various biological data. In this paper, the types of biological data generated by GAN are categorized into two areas: biological sequences and two-dimensional data. These related studies indicated that GANs are able to explore the space of possible data configurations, and tuning the generated data to have specific target properties. This article will provide valuable insights and serve as a starting point for carrying out further studies for researchers.

I. Introduction

As the application of deep learning in biological field increases tremendously, more and more biological data has been processed in various deep learning model. In biology community, various deep learning models have demonstrated their powerful capabilities to discover latent feature and improve accuracy of many prediction tasks, such as microarray data [l], cancer diagnosis and classification[2]. However, due to the high cost of biological data access and the privacy issues, collecting a large amount of biological data for training deep learning model is difficult. For example in medical and some other fields, the sensitivity of the data is high. Concerning this issue, this article focuses on a special type of deep learning model - GANs, and reviews their representative applications for generating biological data.