Using machine learning methods to study the tumour microenvironment and its biomarkers in osteosarcoma metastasis

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Osteosarcoma (OS) metastasis has a poor long-term prognosis.
• There is a critical need for novel biomarkers and innovative research methodologies.

Purpose of the Study:

• To identify potential biomarkers for osteosarcoma metastasis.
• To explore gene expression patterns and their correlation with patient survival and immune infiltration.

Main Methods:

• Utilized data from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) databases.
• Employed synthetic minority oversampling technique (SMOTE) for class imbalance.
• Applied feature selection algorithms (MCFS, Borota, mRMR, LightGBM) and machine learning models (SVM, XGBoost, RF, kNN).
• Used interpretable machine learning (IML) for network construction and analysis of selected genes.

Main Results:

• Identified 53, 45, and 46 features in different datasets using IML.
• Developed 79 interpretable prediction rules for OS metastasis.
• Investigated 39 crucial molecules, finding some act as both predictors and differentially expressed genes.
• Discovered significant survival differences based on TRIP4, S100A9, SELL, and SLC11A1 expression, and their correlation with 22 immune cell types.

Conclusions:

• The identified biomarkers have significant implications for personalized osteosarcoma therapies.
• These findings may enhance the clinical prognosis for patients with osteosarcoma metastasis.