Gene selection for multiclass prediction by weighted Fisher criterion

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression profiling is crucial for understanding disease molecular signatures and developing diagnostics.
• Effective gene selection is vital for accurate disease classification and prediction.
• Existing methods may not optimally identify gene subsets for complex multiclass phenotypes.

Purpose of the Study:

• To propose and evaluate a two-step gene selection method for identifying informative gene subsets.
• To enhance the accuracy of multiclass phenotype classification using selected genes.
• To validate the method's efficacy on real-world microarray datasets.

Main Methods:

• A two-step approach involving individually discriminatory genes (IDGs) and jointly discriminatory genes (JDGs).
• Utilized one-dimensional and multidimensional weighted Fisher criterion (wFC) for gene discrimination.
• Employed artificial neural networks (ANNs) and support vector machines (SVMs) for performance evaluation.

Main Results:

• Successfully identified smaller, efficient gene subsets for diagnosing small round blue cell tumors (SRBCTs) and muscular dystrophies (MDs).
• Achieved high prediction accuracies: 96.9% for SRBCTs and 92.3% for MDs.
• Demonstrated the effectiveness of the IDG/JDG approach in improving multiclass prediction.

Conclusions:

• The proposed two-step gene selection method effectively identifies highly discriminative gene subsets.
• This method enhances the accuracy of multiclass disease prediction.
• The IDG/JDG approach offers a robust strategy for molecular diagnostics.