The identification of informative genes from multiple datasets with increasing complexity.

Area of Science:

• Bioinformatics
• Systems Biology
• Computational Biology

Background:

• Gene regulatory network modeling is challenged by data quality and biological complexity.
• Utilizing multiple datasets from related biological systems can yield more robust models.
• A novel framework for regulatory network modeling using independent training and evaluation datasets was developed.

Purpose of the Study:

• To develop a robust framework for modeling gene regulatory networks.
• To identify informative genes and model interactions across multiple datasets.
• To improve the accuracy and reliability of gene expression prediction.

Main Methods:

• Ordering datasets by noise and informativeness.
• Selecting Bayesian classifiers based on predictive performance on independent data.
• Comparing gene selections and the impact of model complexity.
• Functional analysis of informative genes.

Main Results:

• Identified optimal model complexity for predicting gene expression using cross-validation and independent validation.
• Demonstrated that models trained on simpler datasets can identify gene interactions and informative genes.
• Showed significant improvement in ranking myogenesis-related genes (P < 0.004).
• Outperformed a concordance method in identifying informative genes from multiple datasets of increasing complexity.

Conclusions:

• Bayesian networks from simpler systems outperform those from complex systems.
• Predictive and consistent genes across datasets are fundamentally involved in biological processes.
• Networks trained on simpler systems can model complex interactions in more intricate datasets.