PT  - JOURNAL ARTICLE
AU  - Zhang Zhang
AU  - Lifei Wang
AU  - Shuo Wang
AU  - Ruyi Tao
AU  - Jingshu Xiao
AU  - Muyun Mou
AU  - Jun Cai
AU  - Jiang Zhang
TI  - Neural Gene Network Constructor: A Neural Based Model for Reconstructing Gene Regulatory Network
AID  - 10.1101/842369
DP  - 2019 Jan 01
TA  - bioRxiv
PG  - 842369
4099  - http://biorxiv.org/content/early/2019/11/14/842369.short
4100  - http://biorxiv.org/content/early/2019/11/14/842369.full
AB  - Reconstructing gene regulatory networks (GRNs) and inferring the gene dynamics are important to understand the behavior and the fate of the normal and abnormal cells. Gene regulatory networks could be reconstructed by experimental methods or from gene expression data. Recent advances in Single Cell RNA sequencing technology and the computational method to reconstruct trajectory have generated huge scRNA-seq data tagged with additional time labels. Here, we present a deep learning model “Neural Gene Network Constructor” (NGNC), for inferring gene regulatory network and reconstructing the gene dynamics simultaneously from time series gene expression data. NGNC is a model-free heterogenous model, which can reconstruct any network structure and non-linear dynamics. It consists of two parts: a network generator which incorporating gumbel softmax technique to generate candidate network structure, and a dynamics learner which adopting multiple feedforward neural networks to predict the dynamics. We compare our model with other well-known frameworks on the data set generated by GeneNetWeaver, and achieve the state of the arts results both on network reconstruction and dynamics learning.