Sparse principal component analysis based on genome network for correcting cell type heterogeneity in epigenome-wide...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Epigenome-wide association studies (EWAS) can yield false discoveries due to mixed cell type methylation. Existing non-reference models, like sparse PCA, do not leverage known gene network structures.
• Integrating prior gene network knowledge into EWAS models is an open research challenge to improve accuracy.

Purpose of the Study:

• To introduce GN-ReFAEWAS, a novel non-reference analysis model for EWAS.
• To integrate prior gene network structure into the principal component analysis (PCA) framework to control false discoveries in EWAS.

Main Methods:

• Developed GN-ReFAEWAS, a non-reference analysis model.
• Integrated prior gene network structure into a PCA framework.
• Evaluated the model using one simulated dataset, three real datasets, and three additional tests, comparing it against four existing models.

Main Results:

• GN-ReFAEWAS demonstrated improved performance over existing models.
• The model showed 2-90% better results in sensitivity, specificity, genomic control factor (λ), and correlation coefficient factor (cov) with known cell phenotype ratios.
• Experimental results confirmed the model's effectiveness in controlling false discoveries.

Conclusions:

• GN-ReFAEWAS effectively integrates gene network information into EWAS analysis.
• The proposed model enhances the accuracy and reliability of EWAS by reducing false positive findings.
• This approach offers a significant improvement for identifying phenotype-related methylation sites.