MapReduce-Based Parallel Genetic Algorithm for CpG-Site Selection in Age Prediction.

Area of Science:

• Genomic epigenetics and computational biology.
• Development of efficient algorithms for high-dimensional biological data analysis.

Background:

• DNA methylation (DNAm) serves as a genomic biomarker for predicting human age.
• High-dimensional DNAm data poses computational challenges for predictive modeling.
• Existing methods require significant computational resources and time for feature selection.

Purpose of the Study:

• To propose a novel two-stage parallel algorithm for efficient selection of age-related CpG-sites from DNAm data.
• To reduce the computational time and improve the scalability of age prediction models.
• To enhance the accuracy of age prediction using selected genomic biomarkers.

Main Methods:

• A two-stage parallel algorithm involving data clustering by age range and a MapReduce-based Parallel Genetic Algorithm (MR-based PGA) for feature selection.
• Implementation of a novel parallel framework for data and task parallelization during algorithm execution.
• Utilized 16 healthy human blood DNAm datasets, combined and split into 70% train and 30% test sets for model building and evaluation.

Main Results:

• The proposed method achieved a Mean Absolute Deviation (MAD) of 3.62 years and R² of 95.96% on the unseen test dataset.
• Achieved high accuracy on the train data with MAD of 1.27 years and R² of 99.27%.
• Demonstrated significant computational efficiency: parallelized execution took 58 minutes compared to 4123 minutes for the non-parallelized version.

Conclusions:

• The proposed parallel algorithm is highly efficient and scalable for selecting age-related CpG-sites from high-dimensional DNA methylation data.
• The method offers a substantial reduction in computation time, making age prediction more accessible.
• The approach demonstrates superior performance in age prediction accuracy compared to state-of-the-art methods.