Novel feature selection methods for construction of accurate epigenetic clocks

Abstract

Epigenetic clocks allow the accurate prediction of age based on the methylation status of specific CpG sites in a variety of tissues. These predictive models can be used to distinguish the biological age of an organism from its chronological age, and are a powerful tool to measure the effectiveness of aging interventions. There is a growing need for methods to efficiently construct epigenetic clocks. The most common approach is to create clocks using elastic net regression modelling of all measured CpG sites, without first identifying specific features or CpGs of interest. The addition of feature selection approaches provides the opportunity to reduce the cost and time of clock development by decreasing the number of CpG sites included in clocks. Here, we apply both classic feature selection methods and novel combinatorial methods to the development of epigenetic clocks. We perform feature selection on the human whole blood methylation dataset of [~]470,000 CpG features published by Hannum and colleagues (2015). We develop clocks to predict age, using a variety of feature selection approaches, and all clocks have R2 correlation scores of greater than 0.73. The most predictive clock uses 35 CpG sites for a R2 correlation score of 0.87. The five most frequent sites across all clocks are also modelled to build a clock with a R2 correlation score of 0.83. These two clocks are validated on two external datasets where they maintain excellent predictive accuracy and outperform Hannum et als model in accuracy of age prediction despite using significantly less CpGs. We also identify the associated gene regulatory regions of these CpG sites, which may be possible targets for future aging studies. These novel feature selection algorithms will lower the number of sites needed to be sequenced to build clocks and allow conventionally expensive aging epigenetic studies to cost a fraction of what it would normally.