A Novel Recursive Gene Selection Method Based on Least Square Kernel Extreme Learning Machine.

Area of Science:

• Bioinformatics
• Machine Learning
• Computational Biology

Background:

• Gene selection is crucial for accurate biological data analysis.
• Existing Recursive Feature Elimination (RFE) methods can be computationally intensive.
• Extreme Learning Machine (ELM) classifiers offer efficient learning capabilities.

Purpose of the Study:

• To develop an efficient gene selection method using Least Square Kernel Extreme Learning Machine (LSKELM) and Recursive Feature Elimination (RFE).
• To propose a novel gene importance ranking criterion based on LSKELM weight norms.
• To evaluate the performance of the proposed LSKELM-RFE algorithm against existing RFE methods.

Main Methods:

• Implementation of a Recursive Feature Elimination (RFE) mechanism.
• Utilizing a Least Square Kernel Extreme Learning Machine (LSKELM) classifier for gene analysis.
• Ranking gene importance by the norm of weights derived from LSKELM.
• Iterative removal of least important genes to refine feature set.

Main Results:

• The proposed LSKELM-RFE method effectively selects informative genes.
• LSKELM-RFE demonstrates superior performance compared to two other RFE algorithms.
• The algorithm shows improvements in both computational cost and generalization ability.

Conclusions:

• LSKELM-RFE provides an efficient and effective approach for gene selection.
• The method leverages the random mapping property of ELM kernels, requiring no manual parameter tuning.
• LSKELM-RFE offers a promising alternative for high-dimensional biological data analysis.