Multiple-input multiple-output causal strategies for gene selection.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Traditional variable selection methods in high-dimensional classification focus on relevance, not causation.
• High correlation between variables does not imply a direct causal relationship.
• Existing methods often fail to identify direct causes, limiting biological interpretation.

Purpose of the Study:

• To incorporate causal information into gene selection algorithms.
• To move from single-input single-output to multiple-input multiple-output settings for gene selection.
• To improve both prediction accuracy and biological interpretability in high-dimensional classification.

Main Methods:

• Developed a relevance score incorporating a causal term.
• Applied the method to synthetic datasets for variable prioritization.
• Conducted a meta-analysis of six breast cancer microarray datasets.

Main Results:

• A synthetic case study demonstrated improved prioritization of causal variables.
• The causal approach enhanced prediction accuracy in breast cancer datasets.
• Biological interpretation of results supported the potential of causal gene selection.

Conclusions:

• Integrating causal information into gene selection is effective.
• The approach enhances both prediction accuracy and biological interpretability.
• Causal gene selection offers a more robust method for high-dimensional data analysis.