MSPJ: Discovering potential biomarkers in small gene expression datasets via ensemble learning.

Area of Science:

• Transcriptomics
• Bioinformatics
• Machine Learning

Background:

• Differentially expressed genes (DEGs) are crucial for understanding complex disease pathogenesis.
• Detecting DEGs is challenging in small gene expression datasets due to limitations like sample availability and funding.
• Existing methods often lack the power and stability required for small sample analyses.

Purpose of the Study:

• To introduce MSPJ, a new machine learning approach for robust DEG identification in small transcriptome datasets.
• To improve the power and stability of DEG detection when sample sizes are limited.
• To provide a reliable method for feature selection in transcriptomic analyses.

Main Methods:

• MSPJ utilizes an ensemble learning strategy combining improved multiple random sampling with meta-analysis, Support Vector Machine-Recursive Feature Elimination (SVM-RFE), and permutation testing.
• The method was evaluated using 94 simulated datasets and benchmarked against 10 classical methods on 165 real-world datasets.
• Performance was specifically assessed for small gene expression datasets, particularly those with fewer than 30 samples.

Main Results:

• MSPJ demonstrated superior performance in identifying DEGs across most small gene expression datasets compared to ten classical methods.
• The method showed particular effectiveness in datasets with sample sizes below 30.
• MSPJ provides robust feature selection, enhancing the reliability of DEG identification in low-sample scenarios.

Conclusions:

• MSPJ is an effective machine learning approach for robust DEG identification in small transcriptome datasets.
• The method addresses the limitations of small sample sizes in transcriptomic research.
• MSPJ is expected to advance research into the molecular mechanisms of complex diseases and phenotypes.