IPJGL

IPJGL infers differential gene networks by integrating adaptive gene-importance penalization into Gaussian graphical models to capture condition-specific gene interactions.

Key Features:

• Importance-penalized regularization: Penalizes genes based on relative importance to account for varied mutation probabilities and lower mutation tolerance of essential genes.
• Joint graphical lasso framework: Implements a joint graphical lasso within Gaussian graphical models for differential network inference across conditions.
• APC2 metric: Introduces APC2 (Adaptive Penalized Correlation Coefficient) to evaluate differential levels of gene pairs.
• Validation on simulated and real data: Validated using simulation experiments and analyses of real datasets, including TCGA colorectal and breast cancer datasets.
• Candidate gene identification and survival analysis: Identifies candidate cancer genes with significant survival associations, including examples SOST (colorectal) and RBBP8 (breast).
• Reactome comparison: Compares inferred interactions with interactions documented in the Reactome database.

Scientific Applications:

• Differential network inference: Inferring condition-specific gene regulatory and interaction networks using adaptive importance weighting.
• Cancer gene discovery and prognosis: Detecting candidate cancer genes and assessing survival associations in TCGA colorectal and breast cancer datasets (e.g., SOST, RBBP8).
• Network validation: Benchmarking inferred interactions against Reactome-curated interactions.
• Pairwise differential interaction scoring: Quantifying differential levels of gene pairs with the APC2 metric.

Methodology:

Uses an importance-penalized joint graphical lasso within Gaussian graphical models, computes the APC2 (Adaptive Penalized Correlation Coefficient) for gene-pair differential assessment, and evaluates results via simulation experiments, survival analysis, and comparison to Reactome interactions.