Time Delayed Causal Gene Regulatory Network Inference with Hidden Common Causes.

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Inferring gene regulatory networks (GRNs) is vital for cellular understanding.
• Current computational methods often rely on time series expression data but assume causal sufficiency, ignoring unobserved factors.
• The assumption of causal sufficiency is unrealistic as unobserved variables can influence observed ones.

Purpose of the Study:

• To develop a novel algorithm, HCC-CLINDE, for inferring GRNs from time series data that accounts for the presence of hidden common causes.
• To overcome the restrictive assumption of causal sufficiency in existing GRN inference methods.
• To enable the utilization of multiple short time series for more realistic GRN inference.

Main Methods:

• Developed the HCC-CLINDE algorithm to infer GRNs from time series data, allowing for hidden common causes.
• Modeled GRNs as sparse causal graphs with continuous variables, causal links, and delays.
• Algorithm designed to handle multiple short time series and assumes unobserved variables have observed children and parents with specific connectivity constraints.

Main Results:

• Extensive experiments on synthetic data (up to 100 nodes, 10 hidden nodes) demonstrated HCC-CLINDE's ability to recover the true causal GRN.
• The algorithm showed robustness to minor deviations from Gaussian error distributions.
• Successfully demonstrated potential on small YEASTRACT subnetworks using limited real data.

Conclusions:

• HCC-CLINDE is a novel and less restrictive algorithm for GRN inference compared to previous methods.
• The algorithm effectively infers GRNs in the presence of unobserved common causes.
• HCC-CLINDE shows promise for analyzing real biological data, even with limited time series.