Multi-tissue DNA methylation microarray signature is predictive of gene function.

Area of Science:

• Genomics
• Epigenetics
• Bioinformatics

Background:

• Gene expression correlation networks predict gene function.
• Predictive capacity of DNA methylation at CpG sites for gene function is less understood.
• Guilt-by-association methods can be adapted for DNA methylation analysis.

Purpose of the Study:

• Compare DNA methylation and gene expression in predicting Gene Ontology terms.
• Evaluate co-methylation and machine learning approaches for gene function prediction.
• Assess different methods for generating representative gene methylation values.

Main Methods:

• Utilized Methylation Array Network Analysis (MANA) with over 24,000 samples.
• Generated a network of correlated DNA methylation values.
• Compared probe-level data collapsing methods and genomic regions for correlation matrix generation.

Main Results:

• Mean gene body methylation achieved a prediction macro-AUC of 0.60 for Gene Ontology terms.
• Logistic regression with mean gene body methylation yielded a macro-AUC of 0.82, outperforming a naive predictor (0.72).
• Co-methylated genes were enriched for chromatin state, PRC2, immune response, and development terms.

Conclusions:

• Genes with correlated methylation states exhibit functional relationships.
• Co-methylation analysis is a valuable tool for predicting gene function.
• Epigenetic patterns provide insights into biological processes and gene regulation.