SEMtree

SEmtree identifies functional modules within protein-protein interaction (PPI) networks that exhibit condition-specific changes in molecular activity or phenotypic signatures.

Key Features:

• Tree-Based Structure Learning: Employs tree-based structure discovery algorithms that integrate graph-theoretic and statistical parameters to identify network regions of interest.
• Structural Equation Models Framework: Uses structural equation models (SEM) to integrate condition-specific changes from differential expression and gene-gene co-expression and to enable statistical testing of nodes, directed edges, and paths between groups.
• Implementation: Implemented as the R package SEMgraph.
• Active Subnetwork Detection: Starts from seed genes or gene P-values to generate perturbed modules using five state-of-the-art active subnetwork detection methods.
• Causal Additive Trees and Chu–Liu–Edmonds: Processes perturbed modules through causal additive trees based on the Chu–Liu–Edmonds algorithm to convert undirected edges into directed trees.
• Directed Active Subnetwork Comparison: Converts networks into directed trees to enable comparison of methods in their ability to identify directed active subnetworks.
• Biological Relevance Capture: Identifies biologically relevant subnetworks and directed paths that reflect condition-specific molecular perturbations.
• Performance Evaluation: Demonstrated performance in perturbation extraction and classifier tasks on real datasets such as a COVID-19 RNA-seq dataset and on simulated datasets with varied differential expression patterns.

Scientific Applications:

• Functional module identification in PPI networks: Detection of condition-specific subnetworks to study cellular processes and disease-associated molecular changes.
• Differential activity analysis between conditions: Statistical testing of nodes, directed edges, and paths to compare molecular activity across groups.
• Perturbation extraction and classifier development: Extraction of perturbed modules for downstream classification and predictive modeling tasks.
• Infectious disease and transcriptomic studies: Application to RNA-seq datasets such as COVID-19 for identification of disease-relevant subnetworks.

Methodology:

Applies tree-based structure discovery and SEMs integrating differential expression and gene-gene co-expression; uses seed genes or gene P-values to generate perturbed modules with five active subnetwork detection methods and converts undirected networks into directed trees via causal additive trees using the Chu–Liu–Edmonds algorithm, with statistical testing of nodes, edges, and paths between groups.