Metrics for evaluating differentially methylated region sets predicted from BS-seq data.

Area of Science:

• Epigenetics and Genomics
• Bioinformatics and Computational Biology

Background:

• Differentially methylated regions (DMRs) are crucial for understanding tissue-specific gene expression and serve as biomarkers in noninvasive diagnostics.
• Existing methods for DMR detection lack standardized benchmarks, hindering method selection and downstream application.
• Longer DMRs with more CpG sites are beneficial for biomarker applications.

Purpose of the Study:

• To propose novel metrics, Qn and Ql, for evaluating DMR sets derived from bisulfite sequencing (BS-seq) data.
• To address the challenge of selecting appropriate DMR detection methods and DMR sets for research.
• To provide a reliable method for assessing DMR set quality without requiring additional biological data.

Main Methods:

• Development of two metrics, Qn and Ql, that weight CpG numbers and lengths of DMRs differently.
• Evaluation of eight DMR detection methods using simulated and real BS-seq datasets.
• Comparison of proposed metrics against benchmark metrics and biological data enrichment analyses (genomic features, transcription factors, histones).

Main Results:

• The proposed Qn and Ql metrics demonstrate high correlation with benchmark metrics on simulated data.
• Qn and Ql show strong correlation with biological data enrichment analyses on real BS-seq data.
• The metrics effectively evaluate DMR sets without the need for external biological information.

Conclusions:

• The Qn and Ql metrics offer a robust and data-driven approach for selecting optimal DMR sets from BS-seq data.
• These metrics facilitate the choice of superior DMR detection methods and improve the reliability of DMR-based applications.
• The proposed metrics enhance the utility of DMRs in fields like epigenetics research and noninvasive diagnostics.