High-order dynamic Bayesian Network learning with hidden common causes for causal gene regulatory network.

Area of Science:

• Bioinformatics
• Systems Biology
• Computational Biology

Background:

• Inferring gene regulatory networks (GRNs) is crucial in bioinformatics.
• High-Order Dynamic Bayesian Networks (HO-DBNs) model GRNs with time delays.
• Existing methods often assume causal sufficiency, ignoring unobserved common causes.

Purpose of the Study:

• Develop a computational method to infer GRNs that accounts for hidden common causes.
• Address limitations of previous methods that do not handle multi-step time delays or restrict parent-child relationships of hidden variables.
• Enable GRN inference from multiple, potentially short, time series expression data.

Main Methods:

• Developed a discrete HO-DBN learning algorithm.
• Incorporated assumptions on conditional distribution and relationships of hidden variables (observed children/parents, at least two children, no self-links).
• Algorithm designed to utilize multiple short time series data.

Main Results:

• The algorithm successfully infers hidden common causes in GRNs.
• Demonstrated adequate recovery of causal GRNs from incomplete data using synthetic datasets up to 100 nodes and 10 hidden nodes.
• Showcased potential on limited real expression data from YEASTRACT network.

Conclusions:

• The proposed algorithm effectively infers GRNs in the presence of unobserved common causes.
• Experimental results validate the algorithm's performance on synthetic and real biological data.
• Further validation with more extensive real time series expression data is recommended.