Inference of Genetic Networks From Time-Series and Static Gene Expression Data: Combining a Random-Forest-Based...

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Random-forest-based inference methods offer high performance for analyzing gene expression data.
• Current methods can rank candidate gene regulations but cannot identify those directly affecting a gene of interest.

Purpose of the Study:

• To develop a method for removing unpromising candidate gene regulations.
• To enhance the accuracy of gene regulation confidence values using feature selection.

Main Methods:

• Combined random-forest-based inference with sequential feature selection methods.
• Utilized feature selection outputs to refine confidence scores from the random forest method.

Main Results:

• The combined method improved random-forest performance in 99% of trials on artificial datasets.
• The method successfully removed up to 19% of candidate regulations, with a tendency for small improvements.
• Increased computational cost was observed with the combined approach.

Conclusions:

• The proposed method enhances gene expression data analysis by filtering candidate regulations.
• While improvements are modest and computational cost increases, the method extracts valuable information from limited data.
• Further research is warranted to optimize the balance between improvement and computational efficiency.