Open AccessResearch on Hi-C data enhancement technology based on generative adversarial networks
Author(s) -
Qian Bai,
Zhe Cheng,
Wang Shi-pu,
Wei Zhou
Publication year - 2020
Publication title -
iop conference series. materials science and engineering
Language(s) - English
Resource type - Journals
eISSN - 1757-899X
pISSN - 1757-8981
DOI - 10.1088/1757-899x/782/5/052029
Subject(s) - computer science , resolution (logic) , matrix (chemical analysis) , similarity (geometry) , algorithm , pearson product moment correlation coefficient , measure (data warehouse) , sampling (signal processing) , artificial intelligence , data mining , pattern recognition (psychology) , mathematics , image (mathematics) , statistics , computer vision , chemistry , filter (signal processing) , chromatography
Hi-C technology is one of the most popular tools for studying three-dimensional(3D) genome organization. Due to the high cost of sequencing, most Hi-C data have low resolution and cannot be used to connect distal regulatory elements to their target genes. To solve the problem that hi-c data of high resolution are not easy to obtain, this paper proposes a Hi-C enhancement method (HiCGAN) based on generative adversarial networks, Taking the down-sampling interaction matrix which is highly similar to the original matrix as input, only 1/16 of the original sequencing reading can be used to generate the Hi-C interaction matrix of high resolution. In the experiment, Pearson correlation coefficient was used to measure the similarity between the generated high-resolution matrix and the real high-resolution hi-c matrix in numerical distribution. the apparent interaction pairs were analyzed by Fit-Hi-C, and calling ChromHMM annotates state of 12 kinds of chromatin. Experimental results show that HiCGAN models learned in one cell type can predict high-resolution Hi-C matrices for other cell types. This study proposes a computational framework (HiCGAN) for accurately predicting Hi-C data improving the resolution of Hi-C data.