Optimizing gene selection and module identification via ontology-based scoring and deep learning.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Gene interactions are crucial for understanding biological systems but are challenging to analyze due to complex, high-dimensional omics data.
• Traditional methods struggle with the hierarchical structure of Gene Ontology (GO) terms, causing redundancy and limiting interpretability.
• Deep learning models require biologically relevant input for enhanced predictive performance.

Purpose of the Study:

• To develop an integrated framework for improved gene selection and interaction analysis.
• To leverage GO semantic similarity and regulatory pathway data for enhanced biological insights.
• To address the limitations of traditional methods in handling hierarchical biological data.

Main Methods:

• A novel statistical algorithm ranks genes based on expression scores and GO semantic similarity.
• A deep neural network model identifies gene interaction modules using regulatory pathway data.
• The framework effectively navigates the hierarchical structure of GO terms (directed acyclic graphs) using a feed-forward architecture and back-propagation.

Main Results:

• Demonstrated improved accuracy in gene selection.
• Enhanced discovery of biologically relevant gene interactions.
• Provided valuable insights into complex disease mechanisms through integrated analysis.

Conclusions:

• The integrated framework offers a robust approach to gene selection and interaction analysis.
• This method effectively handles the complexity of biological data and GO hierarchies.
• The findings contribute to a deeper understanding of gene functions and disease pathways.