ClearF: a supervised feature scoring method to find biomarkers using class-wise embedding and reconstruction.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Biomarker discovery relies on effective feature selection methods.
• Existing methods, including information-theoretic approaches, often suffer from long processing times and information loss due to data discretization.
• There is a need for efficient feature scoring methods suitable for continuous-valued data.

Purpose of the Study:

• To propose a novel supervised feature scoring method named ClearF.
• To develop a method that is suitable for continuous-valued data and reduces computation time.
• To validate the effectiveness of ClearF in biomarker detection and classification tasks.

Main Methods:

• ClearF utilizes class-wise low-dimensional embedding and reconstruction error for feature scoring.
• The method computes a compressed representation of each class and the entire dataset.
• Feature scores are derived from reconstruction errors, correlating with information-theoretic measurements.

Main Results:

• ClearF demonstrated higher classification accuracy compared to established methods.
• The proposed method exhibited significantly lower execution times.
• Simulation confirmed the correlation between ClearF scores and information-theoretic measurements.

Conclusions:

• ClearF is an effective and efficient feature scoring method for continuous data.
• The method successfully identified genes highly associated with breast cancer subtypes in TCGA data.
• ClearF offers a promising alternative to existing biomarker detection techniques.