A novel method for cell deconvolution using DNA methylation in PCA space.

Area of Science:

• Genomics
• Bioinformatics
• Epigenetics

Background:

• Reference-based cell deconvolution is crucial for analyzing complex biological samples.
• Existing methods often rely on probe-level data, limiting their scope.
• Principal Component Analysis (PCA) offers a novel feature representation for deconvolution.

Purpose of the Study:

• To develop and validate a novel cell deconvolution method using PCA on DNA methylation array data.
• To compare the performance of the PCA-based method against existing techniques like IDOL-Ext.
• To assess the method's accuracy in diverse datasets, including clinical samples.

Main Methods:

• Feature representation in Principal Component Analysis (PCA) space.
• Application to DNA methylation array data for cell type deconvolution.
• Validation using public datasets, including blood samples from glioma patients.

Main Results:

• The PCA-based method achieves high accuracy (R² 0.73-0.99) comparable to IDOL-Ext.
• Improved discrimination of similar cell types, notably T cell subtypes in glioma samples (R² 0.42-0.75 vs. 0.36-0.66).
• Challenges remain in distinguishing certain cell types like memory CD8 T cells.

Conclusions:

• Cell type deconvolution in PCA space is effective and validated.
• The method demonstrates broad applicability across genomic data types.
• An R package, "BloodCellDecon", is available for public use.