FSR: feature set reduction for scalable and accurate multi-class cancer subtype classification based on copy number

Area of Science:

• Bioinformatics
• Machine Learning
• Genomics

Background:

• High-dimensional data in microarrays poses computational challenges for feature selection.
• Single nucleotide polymorphism (SNP) arrays generate datasets with extremely high feature numbers.
• Scalable feature selection is crucial for accurate disease diagnosis and prognosis.

Purpose of the Study:

• To present a novel feature set reduction (FSR) approach for scalable feature selection.
• To enable efficient handling of high-resolution datasets for improved disease classification.
• To enhance diagnostic and prognostic accuracy for informed clinical decisions.

Main Methods:

• Developed and applied a novel feature set reduction (FSR) approach.
• Evaluated FSR on publicly available cancer SNP array datasets.
• Assessed multiclass predictive classification accuracy, execution speed, and scalability.

Main Results:

• FSR reduced data dimensions by over two orders of magnitude.
• Achieved equal or superior predictive classification performance compared to existing methods.
• Selected features showed strong biological relevance and association with cancer.
• Demonstrated significant speedup and scalability with increasing sample size and array resolution.

Conclusions:

• The proposed FSR approach offers a scalable solution for feature selection in high-dimensional genomic data.
• FSR enhances disease subtype classification accuracy, aiding in diagnosis and prognosis.
• The method effectively identifies biologically relevant features associated with cancer.