基于单细胞RNA测序数据的基因调控网络推断算法综述

Review of gene regulatory network inference algorithms based on single-cell RNA sequencing data

通过基因表达的变化可以推断基因调控网络.单细胞RNA测序(scRNA-seq)为推断细胞周期或分化等时间依赖性生物过程的基因调控网络提供了新的可能性,基于scRNA-seq数据的基因调控网络推断算法成为一个相对活跃的研究方向.本文首先对26种基因调控网络推断算法进行介绍,包括3种针对批量RNA测序数据的推断算法和23种针对scRNA-seq数据的推断算法(基于布尔网络的算法2种、基于微分方程的算法3种、基于伪时序基因相关性集成策略的算法5种、基于共表达基因的算法4种、基于细胞特异性的算法3种、基于深度学习的算法6种),详细描述了每类算法的方法原理和算法优缺点,对算法进行综合比较;然后分析了推断算法比较研究的相关成果,并使用scRNA-seq数据简单评估了26种算法的性能;最后探讨当前基因调控网络推断算法面临的机遇与挑战.

Gene regulatory networks(GRN)can be inferred from the changes of gene expression.Single-cell RNA sequencing(scRNA-seq)technologies provide new possibilities for inferring GRNs of time-dependent biological processes such as cell cycle or differentiation,and GRN inference algorithm has become a relatively active research direction.Firstly,26 inference algorithms including three algorithms based on bulk RNA sequencing data and 23 algorithms based on scRNA-seq data(two algorithms based on Boolean network,three algorithms based on differential equations,five algorithms based on pseudo-time-series gene correlation integration strategy,four algorithms based on co-expression genes,three algorithms based on cell specif-ic,six algorithms based on deep learning)are reviewed.The method principles of the algorithms are described in detail as well as advantages and disadvantages of each algorithm,and the algorithms are compared comprehensively.And then the compara-tive studies on inference algorithms are analyzed,and the performance of 26 algorithms is simply evaluated using scRNA-seq data.Finally,the opportunities and challenges faced by current GRN inference algorithms are discussed.