GeneNet

GeneNet infers causal gene networks from gene expression time series data to distinguish direct from indirect associations and identify directionality in genomic interactions.

Key Features:

• Causal Network Inference: GeneNet employs a heuristic algorithm to infer directed graphical models from high-dimensional genomic data, emphasizing causal relationships rather than correlations.
• Partial Correlation Graphs: The method converts correlation networks into partial correlation graphs to identify direct associations between genes.
• Node Ordering and Testing: A partial ordering of nodes is established via multiple testing of the log-ratio of standardized partial variances to enable construction of directed acyclic causal networks as subgraphs.
• Efficiency and Approximation: The heuristic substitutes lower-order partial correlations with full-order ones to approximate causal structures efficiently, particularly in small sample sizes and sparse networks.
• Application Example: The methodology has been applied to an Arabidopsis thaliana expression dataset.

Scientific Applications:

• Causal gene network discovery: Infer directed acyclic graphs from time-series expression data to reveal causal regulatory interactions.
• Genetic regulatory mechanism analysis: Identify directionality of interactions relevant to regulatory mechanisms.
• Pathway analysis: Distinguish direct from indirect gene associations to support pathway-level interpretation.
• High-dimensional, small-sample studies: Apply to high-dimensional genomic datasets with limited sample sizes and sparse network structure.

Methodology:

Computational steps explicitly include conversion of correlation networks into partial correlation graphs; multiple testing of the log-ratio of standardized partial variances to establish node ordering; identification of a directed acyclic causal network as a subgraph of the partial correlation network; and substitution of lower-order partial correlations with full-order ones for computational efficiency.