Transcriptional network classifiers.

Area of Science:

• Genomics
• Systems Biology
• Bioinformatics

Background:

• Gene interactions are crucial for transcriptional networks and disease processes.
• Existing methods for network reconstruction often lack accuracy in disease classification.
• A systems biology approach is needed to infer function-dependent networks for tissue state identification.

Purpose of the Study:

• To develop a systems biology approach for inferring function-dependent transcriptional networks.
• To utilize these networks as classifiers for identifying tissue states and improving disease diagnosis.
• To achieve higher classification accuracy compared to existing methods by considering gene interactions.

Main Methods:

• Employs the Bayesian networks framework for network inference.
• Incorporates a two-step algorithm: gene filtering by Bayes factor and collinearity elimination via network learning.
• Validates the approach using two clinical datasets: lung cancer subtypes and thoracic aortic aneurysm (TAA).

Main Results:

• A 25-gene classifier for lung cancer subtypes achieved 95% accuracy on independent samples.
• A 34-gene classifier for TAA diagnosis reached 82% accuracy on independent samples.
• The proposed method demonstrated superior classification accuracy and more compact gene signatures compared to PCA/LDA, PAM, and Weighted Voting.

Conclusions:

• The systems biology approach effectively infers function-dependent transcriptional networks for accurate biological sample classification.
• Clinical data validation confirms the approach's potential for disease diagnosis.
• The method offers a promising tool for identifying tissue states and aiding in medical diagnostics.