Framework for classification of cancer gene expression data using Bayesian hyper-parameter optimization.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Cancer classification from gene expression data is challenging due to high dimensionality, limited samples, and class imbalance.
• Existing methods often struggle with the complexity and scale of genomic datasets.

Purpose of the Study:

• To develop and evaluate a robust computational framework for accurate cancer gene expression profile classification.
• To enhance diagnostic efficiency by optimizing feature selection and classifier performance.

Main Methods:

• A multi-stage pipeline involving data pre-processing (scaling, normalization), recursive feature elimination (RFE) with genetic algorithm refinement, and a diverse classifier meta-pool.
• Hyper-parameter optimization using Bayesian Optimization and a novel algorithm for selecting the best classifier based on accuracy and computation time.

Main Results:

• The proposed framework demonstrated significant improvements in classification accuracy and reduced computation time on multiple microarray and PAN-Cancer RNA sequencing datasets.
• The selected classifier consistently outperformed state-of-the-art methods in predicting cancer types.

Conclusions:

• The developed framework offers a superior approach to computational cancer classification using gene expression data.
• This method provides a more efficient and accurate tool for cancer diagnosis and research.