BoostDiff

BoostDiff infers differential gene regulatory networks by leveraging boosted differential regression trees to detect network rewiring between two biological conditions.

Key Features:

• Boosted differential regression trees: Uses an ensemble of regression trees that explicitly model differences in regulatory interactions between two conditions.
• AdaBoost (Adaptive Boosting): Constructs a series of differential trees via an AdaBoost ensemble to enhance detection of subtle interaction changes.
• "Differential variance improvement" splitting criterion: Employs the novel "differential variance improvement" metric as the tree split criterion to prioritize splits that reveal differential regulatory signals.
• Variable importance measures: Derives variable importance from the ensemble to quantify alterations in gene expression predictability across conditions.
• Differential network construction: Builds differential networks from importance measures to highlight significant shifts in regulatory interactions and context-specific networks enriched with disease-relevant pathways.
• Validation and benchmarking: Demonstrated superior performance on simulated data across four complexity settings and validated on real transcriptomics datasets.

Scientific Applications:

• Network rewiring detection: Identification of complex changes in gene regulatory interactions between two biological conditions.
• COVID-19 transcriptomics: Application to COVID-19 datasets to identify context-relevant differential networks.
• Crohn's disease transcriptomics: Application to Crohn's disease datasets to reveal disease-associated network changes.
• Breast cancer transcriptomics: Application to breast cancer datasets to detect cancer‑associated regulatory rewiring.
• Prostate adenocarcinoma transcriptomics: Application to prostate adenocarcinoma datasets to characterize tumor-specific network alterations.
• Bacillus subtilis stress response: Application to B. subtilis transcriptomics under stress conditions to identify microbial stress‑response network changes.

Methodology:

Uses an AdaBoost ensemble to construct differential regression trees, applies the "differential variance improvement" splitting criterion, and derives variable importance measures from the ensemble to quantify alterations and construct differential networks.