A Fast Reference-Free Genome Compression Using Deep Neural Networks
Zeinab Nazemi Absardi, Reza Javidan · 2019
Recent development of DNA sequencing technologies has led to a significant increase in genomic data volume. Such a big amount of genome data needs appropriate data storage, data management, and data transfer policies. Compressing genomes can be used for efficient data management. Auto-encoder is a kind of deep neural networks, due to its ability to reduce the dimension of data is suitable for this purpose. In this paper, a new method for genome compression with auto-encoders based on deep neural networks is proposed. It is the first time that an auto-encoder is used to compress the genomes. Experimental results showed that the proposed method can achieve a compression ratio of up to 5 and 92 percent compression accuracy in case of reference-free genome compression. Moreover, after the auto-encoder training stage, the trained network will have a very short compression time which makes it suitable for real-time applications.