FS-GBDT: identification multicancer-risk module via a feature selection algorithm by integrating Fisher score and...

Area of Science:

• Genomics
• Bioinformatics
• Cancer Research

Background:

• Cancer is a complex, heterogeneous disease with shared underlying mechanisms across different types.
• Identifying critical genes is essential for understanding cancer development and progression.
• Joint analysis of multiple cancers can reveal overlapping oncogenic pathways.

Purpose of the Study:

• To develop a robust feature selection framework for identifying key genes in high-dimensional cancer gene expression data.
• To explore overlapping cancer mechanisms by analyzing 11 human cancer types.
• To validate the proposed framework against existing methods and analyze the biological significance of identified genes.

Main Methods:

• Proposed a fusion feature selection framework: Fisher score and Gradient Boosting Decision Tree (FS-GBDT).
• Conducted a joint analysis of gene expression data from 11 human cancer types.
• Compared FS-GBDT with four other feature selection algorithms using a Support Vector Machine classifier.
• Performed Gene Ontology analysis and literature validation on the selected key genes.

Main Results:

• The FS-GBDT framework successfully identified a robust subset of key feature genes across multiple cancer types.
• FS-GBDT significantly outperformed four other common feature selection algorithms in terms of classification accuracy.
• The identified key genes were organized into functional modules, offering potential as disease markers.

Conclusions:

• FS-GBDT is an effective method for selecting decisive feature genes in complex, high-dimensional cancer datasets.
• The identified functional modules represent core cancer mechanisms and can serve as reliable disease markers.
• Joint analysis of multiple cancers facilitates the discovery of shared genetic drivers and therapeutic targets.