GENIE3

GENIE3 infers gene regulatory networks from high-throughput genomic data, particularly microarray gene expression datasets, to identify regulatory interactions among genes.

Key Features:

• Decomposition into Regression Problems: Breaks the GRN inference problem for p genes into p regression problems by predicting each target gene's expression from the expression of all other genes.
• Tree-Based Ensemble Methods: Employs tree-based ensemble methods, specifically Random Forests and Extra-Trees, to model relationships and capture complex non-linear interactions.
• Feature Importance for Network Inference: Uses feature importance scores from the ensemble models to indicate potential regulatory links and to rank interactions.
• Directed Networks: Produces directed gene regulatory networks by assigning regulator-to-target relationships based on prediction importance.
• Assumption-Free and Versatile: Makes no specific assumptions about the form of gene regulation and can handle combinatorial and non-linear interactions.
• Efficiency and Scalability: Designed for speed and scalability to accommodate large datasets.
• Extensibility: Can be applied beyond synthetic benchmarks to real biological datasets and extended to other types of genomic data.

Scientific Applications:

• Performance Evaluation: Evaluated in the DREAM4 In Silico Multifactorial challenge, where it ranked among top-performing GRN inference algorithms on simulated data.
• Real Data Application: Applied to reconstruct the genetic regulatory network of Escherichia coli from expression data.

Methodology:

Decomposes network inference into p regression problems predicting each target gene from all other genes, models each regression with Random Forests or Extra-Trees, and uses the resulting feature importance scores to rank interactions and reconstruct a directed GRN.