东北大学学报(自然科学版) ›› 2023, Vol. 44 ›› Issue (3): 323-330.DOI: 10.12068/j.issn.1005-3026.2023.03.003

• 信息与控制 • 上一篇    下一篇

基于潜在调控因子筛选的高阶动态贝叶斯建模方法

李婵1, 曲璐渲1, 信俊昌2,3, 王之琼1   

  1. (1. 东北大学 医学与生物信息工程学院, 辽宁 沈阳110169; 2. 东北大学 计算机科学与工程学院, 辽宁 沈阳110169; 3. 辽宁省大数据管理与分析重点实验室, 辽宁 沈阳110169)
  • 修回日期:2022-01-14 接受日期:2022-01-14 发布日期:2023-03-24
  • 通讯作者: 李婵
  • 作者简介:李婵(1994-),女,河北邯郸人,东北大学硕士研究生; 信俊昌(1977-),男,辽宁辽阳人,东北大学教授,博士生导师; 王之琼(1980-),女,黑龙江哈尔滨人,东北大学教授,博士生导师.
  • 基金资助:
    国家自然科学基金资助项目(62072089); 中央高校基本科研业务费专项资金资助项目(N2116016, N2104001, N2019007, N2224001-10).

High-Order Dynamic Bayesian Network Modelling Method Based on Potential Regulatory Factors Screening

LI Chan1, QU Lu-xuan1, XIN Jun-chang2,3, WANG Zhi-qiong1   

  1. 1. College of Medicine & Biological Information Engineering, Northeastern University, Shenyang 110169, China; 2. School of Computer Science & Engineering, Northeastern University, Shenyang 110169, China; 3. Key Laboratory of Big Data Management and Analytics
  • Revised:2022-01-14 Accepted:2022-01-14 Published:2023-03-24
  • Contact: WANG Zhi-qiong
  • About author:-
  • Supported by:
    -

摘要: 为了解决目前用于构建基因调控网络的方法中所存在的网络构建准确率低、网络构建时间过长等问题,以及减小网络构建的复杂度,提高网络构建效率,提出了一种基于潜在调控因子筛选的高阶动态贝叶斯网络建模方法(high-order dynamic Bayesian network modeling method based on potential regulatory factor screening, PRS-HO-DBN).该方法将关联模型与高阶动态贝叶斯网络模型相结合,首先利用潜在调控因子筛选的方法在不同的时间延迟下删除与目标基因关联程度较低的基因,保留与目标基因关联程度较高的基因并作为目标基因的潜在调控因子集,以减小搜索空间;然后利用高阶动态贝叶斯模型进行结构学习,以提高网络构建的精确率.与其他的网络构建模型方法相比,该方法可以极大地缩短网络构建的时间,提升效率和精确度.

关键词: 基因调控网络;潜在调控因子;高阶动态贝叶斯网络;关联模型;结构学习

Abstract: In order to solve the problems of low network construction accuracy and long network construction time in the current methods used to construct gene regulatory networks, so as to reduce the complexity of network construction and improve the efficiency of network construction, a method called high-order dynamic Bayesian network modelling method based on potential regulatory factors screening (PRS-HO-DBN) was proposed. The method combines the correlation model with the high-order dynamic Bayesian network model. Firstly, the potential regulatory factor screening method is used to delete the genes with low association with the target gene under different time delays, and retain the genes with high association with the target gene as the potential regulatory factor set of the target gene to reduce the search space. Then the high-order dynamic Bayesian model is used for structure learning to improve the accuracy of network construction. Compared with other methods, the method can greatly reduce network construction time and improve efficiency and accuracy.

Key words: gene regulatory networks; potential regulatory factors; high-order dynamic Bayesian network; correlation model; structure learning

中图分类号: