Pan-tissue methylation aging clock: Recalibrated and a method to analyze and interpret the selected features

https://doi.org/10.1016/j.mad.2022.111676Get rights and content

Highlights

  • The pan tissue methylation aging clock was recalibrated with a high prediction accuracy using machine learning model

  • The aging CpGs showed a constant change in the methylation level upon aging starting from adult phase.

  • Pathway enrichment analysis showed the enrichment of several development related pathways

  • This phenomena was explained using the antagonistic pleiotropy theory.

Abstract

The abundance of the biological data and the rapid evolution of the newer machine learning technologies have increased the epigenetics research in the last decade. This has enhanced the ability to measure the biological age of humans and different organisms via their omics data. DNA methylation array data are commonly used in the prediction of methylation age. Horvath clock has been adopted in various aging studies as a DNA methylation age predicting clock due to its higher accuracy and multi tissue prediction potential. In the current study, we have developed a pan tissue methylation-aging clock by using the publicly available illumina 450k and EPIC array methylation datasets. In doing that, we developed a highly accurate epigenetic clock, which predicts the age of multiple tissues with higher accuracy. We have also analyzed the selected probes for their biological relevance. Upon analyzing the selected features further, we found out evidences, which support the Antagonistic pleiotropy theory of aging.

Introduction

Aging is an unavoidable process, which takes place in all the living organisms. It is a decline of physiological processes in a time dependent manner. In common the number of years from the time of birth is being used as a measurement of age (Kalache et al., 2002). The rate of aging differs from one organism to another and also in humans, people with the same age shows varied physical and health changes (Baker and Sprott, 1988). It is reasonable to estimate that DNA methylation could potentially be one of the factors to influence the rate of aging (Ferrucci et al., 2020). Therefore, the past decade has seen various researches trying to establish aging markers based on epigenetic changes. One such research is the establishment of clock based on the methylation pattern. This was achieved by the usage of machine learning algorithms. In the year 2013 Hannum developed a methylation age clock (Hannum et al., 2013) using a blood derived dataset with a higher prediction accuracy. However, the model has suffered when it was applied onto other tissues. In the same year Steve Horvath developed a methylation clock using multiple tissues with a near perfect accuracy and was known as DNAmage (Horvath, 2013). In the later years, several different epigenetics clocks were developed with as low as just 3 CpG sites (Lin et al., 2016). In order to predict the lifespan along with the DNA methylation data Levine et al. included blood derived clinical markers and chronological age to develop the DNA PhenoAge (Levine et al., 2018). GrimAge (Lu et al., 2019), another clock model which utilizes smoking habit and plasma protein level was developed to predict the lifespan and the all-cause mortality. Both the clocks were better at predicting the lifespan and mortality factors when compared with the Horvath’s clock, which only utilizes the methylation data.

The recent development of the array technology introduced more target CpG sites and the latest addition to the illumina array is the EPIC methylation array beadchip which covers over 850 K CpG sites. In our study, we wanted to include a broad range of DNA methylation sites. We assumed, this inclusiveness could cover more DNA regions, which may involve in the aging process and results in identifying more important methylation sites during the feature selection. Recently deep learning models have been showing promising performance in various fields, implying that technology in age prediction will yield a better predictive accuracy. Deep learning derived clock model has also been recently published (Galkin et al., 2021b).

In the present study, we explored the potential of the advancement in the array technology to develop a DNA methylation clock that could predict the age of different tissues. In the process, we have developed a model with a higher accuracy and lesser error in the test data when compared with the already existing multi tissue clocks. In addition, this model have performed better in predicting the age of the tissues, which have not been used in the training session. Further, we have also analyzed the probes selected by the machine learning model to find out the biological meaning behind these DNA methylation age probes as they undeniably capture the essential characteristics of aging epigenome (Horvath and Raj, 2018).

Section snippets

Dataset collection

All the datasets used in this study have been collected from the public database Gene Expression Omnibus. In Fig. 1 the study process have been graphically explained. If the original study of the datasets includes any disease samples, they were neglected and only the control samples from the datasets has been used. Therefore, all the data used in this study were of normal humans. We used 20 different tissues datasets, which comprises 4671 samples for the training and test. The list of datasets

Performance of the model in training and test data

To verify the model’s performance, we used various metrics like Root Mean Squared Error (RMSE), Mean Absolute Error (MAE) and the correlation coefficient R2 values, which were measured between the actual age and the predicted age (Fig. 2A and C). The R2 value explains the proportion of the variance. RMSE and MAE were used to measure the average error in the prediction. Table 3 shows the performance values of the clock model in both test and training set data. The R2 value for the training data

Discussion

Numerous distinct epigenetic changes occur with aging (Kane and Sinclair, 2019). In this study, we have developed a machine-learning model, which could accurately predict the age of a person based on the DNA methylation levels. Though several studies developed epigenetic aging clock models, this study is only the second to use multiple tissues to generate a model next to the Horvath clock and the first to utilize a higher number of DNA methylation probes. However it have been proven that the

Conclusions

In this study, we have designed an epigenetic clock using the DNA methylation levels of multiple tissues and analyzed the selected probes for their biological function on aging. As a, result we trained a clock model with higher accuracy in predicting the age of a person. During the early adult phase, the methylation levels were increasing at a constant pace (hyper methylate) and slows down after the age of 80. This shows the evident influence of DNA methylation on aging. We analyzed these

CRediT authorship contribution statement

KAV: Conceptualization, Formal analysis, Data curation, Visualization, Writing – original draft, Writing – review & editing. GWC: Conceptualization, Funding acquisition, Supervision, Data curation, Formal analysis, Visualization, Writing – review & editing.

Competing interest statement

The authors declare that they have no competing interests.

Acknowledgments

We thank all the researchers who made all the microarray data that have been used in this study available for the scientific community. We also thank Dr. Shivarama holla kayyar and Ruban Kumar for their opinions on computing. This study was supported by research fund from Chosun University, 2021.

References (33)

  • J.P. Fortin et al.

    Preprocessing, normalization and integration of the Illumina HumanMethylationEPIC array with minfi

    Bioinformatics

    (2017)
  • F. Galkin et al.

    Aging Dis.

    (2021)
  • F. Galkin et al.

    DeepMAge: a methylation aging clock developed with deep learning

    Aging Dis.

    (2021)
  • P. Garagnani et al.

    Methylation of ELOVL2 gene as a new epigenetic marker of age

    Aging Cell

    (2012)
  • G. Gasparoni et al.

    DNA methylation analysis on purified neurons and glia dissects age and Alzheimer's disease-specific changes in the human cortex

    Epigenetics Chromatin

    (2018)
  • S. Horvath

    DNA methylation age of human tissues and cell types

    Genome Biol.

    (2013)
  • View full text