RF_Purify: a novel tool for comprehensive analysis of tumor-purity in methylation array data based on random forest...

Area of Science:

• Epigenetics
• Computational Biology
• Oncology

Background:

• Array-based methylation analysis is common for tumor studies.
• Bulk tumor samples contain mixed cell populations (tumor, immune, stromal).
• Accurate tumor purity assessment is crucial but challenging without control samples.

Purpose of the Study:

• To develop a novel, reference-free method for quantifying tumor purity.
• To apply and validate this method on brain tumor datasets.
• To enable reliable methylome analysis across diverse tumor types.

Main Methods:

• Developed Random Forest classifiers trained on ABSOLUTE and ESTIMATE purity values from TCGA data.
• Validated classifiers on independent datasets and compared them with existing methods (ESTIMATE, LUMP).
• Applied the model to Illumina methylation array data from brain tumors.

Main Results:

• Random Forest classifiers accurately predict tumor purity using methylation array data.
• The method performs well on datasets not previously characterized for purity.
• Brain tumor subgroups exhibit significant variations in tumor purity.

Conclusions:

• Random Forest-based tumor purity prediction is effective for novel methylation datasets.
• This method does not require prior knowledge of tumor type or matching control tissue.
• Enables robust tumor purity estimation for broader methylome research.