BOOTSTRAP INFERENCE FOR NETWORK CONSTRUCTION WITH AN APPLICATION TO A BREAST CANCER MICROARRAY STUDY.

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Gaussian Graphical Models (GGMs) are utilized for constructing genetic regulatory networks.
• Network inference in high-dimension-low-sample-size scenarios often requires regularization techniques.
• Traditional methods for selecting regularization parameters (e.g., BIC, cross-validation) are often ineffective in unsupervised settings.

Purpose of the Study:

• To propose a new method, Bootstrap Inference for Network COnstruction (BINCO), for inferring genetic regulatory networks.
• To address the challenge of selecting appropriate regularization in unsupervised network inference.
• To directly control the false discovery rates (FDRs) of selected network edges.

Main Methods:

• BINCO infers networks by directly controlling the false discovery rates (FDRs) of selected edges.
• The method employs model aggregation to calculate edge selection frequencies.
• A mixture model is fitted to the distribution of edge selection frequencies to estimate FDRs.

Main Results:

• The proposed BINCO method was applied to infer a gene regulatory network from microarray expression breast cancer data.
• High-confidence edges and well-connected hub genes were identified.
• These findings suggest potential roles in understanding breast cancer's underlying biological processes.

Conclusions:

• BINCO offers a robust approach for network inference by controlling FDRs, particularly in high-dimensional, low-sample-size data.
• The method is broadly applicable beyond genetic network construction.
• Identified network features in breast cancer data provide insights into disease mechanisms.