Robust biomarker screening from gene expression data by stable machine learning-recursive feature elimination...

Area of Science:

• Computational biomedicine
• Bioinformatics
• Machine learning in genomics

Background:

• Identifying robust biomarkers from gene expression data is crucial for diagnosing complex diseases.
• Early detection of spontaneous preterm birth (SPTB) and high-grade serous ovarian cancer (HGSOC) can significantly improve patient outcomes.
• Existing biomarker discovery methods may lack robustness when applied to high-throughput omics data.

Purpose of the Study:

• To propose a stable machine learning-recursive feature elimination (StabML-RFE) strategy for robust biomarker screening.
• To enhance the reliability and accuracy of biomarker identification from gene expression profiling.
• To provide a versatile pipeline for discovering diagnostic biomarkers for complex diseases.

Main Methods:

• Employed eight machine learning algorithms (AdaBoost, Decision Tree, GBDT, Naive Bayes, NNET, Random Forest, SVM, XGBoost) with recursive feature elimination (RFE).
• Developed a stability metric combining classification performance on test data to assess feature selection robustness.
• Selected high-frequency features from stable subsets identified by StabML-RFE as robust biomarkers.

Main Results:

• The StabML-RFE strategy successfully identified robust biomarkers from gene expression data.
• Internal validation, functional enrichment analysis, and literature checks confirmed the screened biomarkers.
• External validation on real-world SPTB and HGSOC datasets demonstrated the pipeline's effectiveness.

Conclusions:

• The proposed StabML-RFE strategy offers a robust method for biomarker discovery from high-throughput gene expression data.
• This approach is applicable to various complex diseases, facilitating early detection and intervention.
• The StabML-RFE pipeline provides a valuable tool for computational biomedicine and personalized medicine.