A systematic evaluation of cell-type-specific differential methylation analysis in bulk tissue

Area of Science:

• Epigenetics
• Computational Biology
• Genomics

Background:

• Bulk DNA methylation data analysis presents challenges in discerning cell-type-specific methylation changes.
• Accurate identification of cell-type-specific differential methylation CpGs is crucial for understanding disease mechanisms.

Purpose of the Study:

• To systematically assess and compare the performance of five computational models (CellDMC, TCA, HIRE, TOAST, CeDAR) for detecting cell-type-specific differential methylation CpGs.
• To evaluate model performance using simulations and real-world epigenome-wide association studies (EWAS) data for rheumatoid arthritis and major depressive disorder.
• To propose and validate methods for integrating multiple model outputs to enhance detection accuracy.

Main Methods:

• Systematic performance evaluation of CellDMC, TCA, HIRE, TOAST, and CeDAR using simulated datasets.
• Application of models to EWAS data from rheumatoid arthritis and major depressive disorder cohorts profiled on Illumina DNA Methylation BeadArrays.
• Development and testing of p-value aggregation strategies, including minimum p-value ($minpv$) and average p-value ($avepv$) approaches.

Main Results:

• Computational models exhibited variable performance based on metrics, sample size, and computational efficiency.
• The proposed $minpv$ and $avepv$ aggregation methods demonstrated significant improvements in identifying cell-type-specific differential methylation CpGs.
• Case studies on rheumatoid arthritis and major depressive disorder highlighted the practical utility and limitations of the evaluated models.

Conclusions:

• No single computational model universally outperformed others across all scenarios.
• Integrating results from multiple models via p-value aggregation offers a robust strategy for enhancing the discovery of cell-type-specific methylation differences.
• This systematic evaluation and proposed integration approach provide valuable guidance for researchers analyzing bulk DNA methylation data in complex diseases.