XGBoost

XGBoost predicts tumor purity from RNA-seq gene expression data using supervised gradient-boosted decision trees.

Key Features:

• Tumor purity definition: Tumor purity is defined as the percentage of cancer cells within a tumor sample and serves as the target variable.
• Data source: Trained and tested on RNA-seq gene expression data from 33 tumor types in The Cancer Genome Atlas (TCGA).
• Model type: Supervised gradient-boosted decision tree models implemented with XGBoost applied to gene expression features.
• Performance metrics: Median correlations between observed and predicted tumor purities ranged from 0.75 to 0.87, with low root mean square errors reported.
• Predictive gene set: A ten-gene set (CSF2RB, RHOH, C1S, CCDC69, CCL22, CYTIP, POU2AF1, FGR, CCL21, and IL7R) consistently predicts tumor purity across tumor types.
• Validation: The ten-gene set was validated on an independent TCGA dataset, with gene expression showing a strong correlation with observed tumor purities (ρ = 0.88).

Scientific Applications:

• Tumor purity estimation: Predicting the percentage of cancer cells across diverse tumor types in TCGA.
• Biomarker identification: Deriving a ten-gene signature as potential biomarkers for tumor purity.
• Tumor microenvironment analysis: Informing studies of the interplay between malignant and non-malignant cell populations in tumor biology.
• Cancer research and diagnostics: Improving accuracy of downstream analyses and diagnostic interpretation by accounting for tumor purity.

Methodology:

Trained supervised XGBoost models on RNA-seq gene expression from 33 TCGA tumor types, evaluated predictions using correlation and root mean square error, identified a ten-gene predictive set, and validated it on an independent TCGA dataset (ρ = 0.88).