Integrated Theory- and Data-driven Feature Selection in Gene Expression Data Analysis

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-dimensional biological data is rapidly increasing, demanding automated knowledge discovery.
• Current theory-driven and data-driven methods have limitations, being biased toward existing knowledge.
• Pure data-driven approaches struggle with the high dimensionality of big biological data.

Purpose of the Study:

• To develop an integrated method addressing the challenges of high-dimensional biological data.
• To present a novel two-step analytical workflow for automated knowledge production.
• To improve the discovery of causal relationships from complex biological datasets.

Main Methods:

• A novel two-step analytical workflow was developed.
• The first step incorporates a new feature selection paradigm for high-throughput gene expression data.
• The second step utilizes graphical causal modeling for automatic extraction of causal relationships.

Main Results:

• The integrated method effectively addresses high dimensionality in big biological data.
• The approach intelligently selects relevant genes for causal network learning.
• Validation was performed on real-world clinical datasets from The Cancer Genome Atlas (TCGA).

Conclusions:

• An integrated approach combining feature selection and graphical causal modeling offers a powerful solution for big biological data analysis.
• This method overcomes the biases of traditional approaches by incorporating comprehensive knowledge.
• The workflow facilitates the intelligent selection of genes and the discovery of causal networks in cancer genomics.