L_2,1-Extreme Learning Machine: An Efficient Robust Classifier for Tumor Classification.

Area of Science:

• Bioinformatics
• Machine Learning
• Genomics

Background:

• Cancer research generates vast gene expression and single-cell RNA sequencing (scRNA-seq) datasets.
• Classifying high-dimensional biological data presents significant technical challenges.
• Traditional Extreme Learning Machine (ELM) methods lack robustness against noise and outliers.

Purpose of the Study:

• To propose a novel, robust Extreme Learning Machine (ELM) method for high-dimensional data classification.
• To improve the robustness of ELM by introducing the L2,1-norm into the loss function.
• To evaluate the efficacy of the proposed method on benchmark datasets and real-world cancer data.

Main Methods:

• Development of the L2,1-Extreme Learning Machine (L2,1-ELM) by incorporating the L2,1-norm into the ELM loss function.
• Evaluation of L2,1-ELM on five UCI benchmark datasets.
• Application and validation of L2,1-ELM on The Cancer Genome Atlas (TCGA) and scRNA-seq datasets for cancer sample classification.

Main Results:

• The proposed L2,1-ELM achieved competitive results on UCI datasets.
• Experimental results on TCGA and scRNA-seq datasets demonstrated the method's effectiveness in cancer sample classification.
• The L2,1-norm effectively minimized the influence of noise and outliers, enhancing classification robustness.

Conclusions:

• The L2,1-Extreme Learning Machine (L2,1-ELM) offers a robust approach for classifying high-dimensional biological data.
• The method shows significant potential for applications in cancer research, particularly in analyzing TCGA and scRNA-seq data.
• ELM and its variants are promising tools for advancing cancer sample classification and analysis.