Predicting DNA methylation from genetic data lacking racial diversity using shared classified random effects.

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Public genomic data lacks racial and ethnic diversity, hindering research.
• Epigenetic alterations, like DNA methylation, are crucial in cancer but understudied due to data limitations.
• Genetic data is more abundant than epigenetic data in public repositories.

Purpose of the Study:

• To develop a model-based framework for predicting DNA methylation from genetic data in racially sparse repositories.
• To improve the accuracy of epigenetic predictions for underrepresented racial groups.
• To aid cervical cancer research using The Cancer Genome Atlas (TCGA) data.

Main Methods:

• A novel prediction approach using shared random effects from a nested error mixed effects regression model.
• Borrowing strength across racial groups to enhance predictive accuracy.
• Comparing the proposed method against elastic net and random forest prediction models.

Main Results:

• The shared classified random effects approach demonstrated uniformly more accurate DNA methylation predictions.
• Improved predictive accuracy was observed overall and for each individual racial group.
• The method effectively leverages genetic data to predict epigenetic alterations in sparse datasets.

Conclusions:

• The developed framework accurately predicts DNA methylation from genetic data, even with limited racial diversity.
• This approach offers a valuable tool for cancer research, particularly for underrepresented populations.
• The methodology enhances the utility of existing genomic data for epigenetic studies.