Reverse engineering module networks by PSO-RNN hybrid modeling.

Area of Science:

• Systems Biology
• Genomics
• Bioinformatics

Background:

• Gene regulatory network (GRN) inference is challenging due to high dimensionality, complex gene relationships, data noise, and small sample sizes.
• Understanding gene interactions is crucial for biological processes like differentiation, cell cycle, and development.

Purpose of the Study:

• To enhance gene interaction understanding and GRN inference accuracy.
• To develop a novel method integrating prior biological knowledge and multiple data sources.
• To decompose complex GRN inference into smaller, manageable network modules.

Main Methods:

• A novel GRN inference method based on a module network model.
• Module selection using fuzzy c-mean (FCM) clustering with gene functional category information.
• Network inference between modules using a hybrid Particle Swarm Optimization-Recurrent Neural Network (PSO-RNN) approach.

Main Results:

• The method was tested on rat central nervous system development and yeast cell cycle data.
• Results were evaluated against previous studies and Gene Ontology annotations.
• The approach successfully identified biologically meaningful modules and networks.

Conclusions:

• The proposed method addresses challenges in GRN inference from time-course gene expression data with limited time points.
• Preprocessing, module identification, and PSO-RNN modeling reduce noise and capture complex gene relationships.
• The study demonstrates the generation of biologically meaningful modules and networks for improved GRN reconstruction.