dnaMethyAge

A user friendly R package to predict epigenetic age and calculate age acceleration from DNA methylation data. Currently, supported age clocks are listed below:

Name	Published Name	First Author (Published Year)	Trained Phenotype	Num. of CpGs	Tissues Derived
HannumG2013		Gregory Hannum (2013)	Chronological age	71	Whole blood
HorvathS2013	Multi-tissue age estimator	Steve Horvath (2013)	Chronological age	353	27 tissues/cells
YangZ2016	epiTOC	Zhen Yang (2016)	Mitotic divisions	385	Whole blood
ZhangY2017		Yan Zhang (2017)	Mortality risk	10	Whole blood
HorvathS2018	Skin & Blood Clock	Steve Horvath (2018)	Chronological age	391	Skin, blood, buccal cells and 5 other tissues
LevineM2018	PhenoAge	Morgan E. Levine (2018)	Mortality risk	513	Whole blood
McEwenL2019	PedBE	Lisa M. McEwen (2019)	Chronological age	94	Buccal epithelial cells
ZhangQ2019		Qian Zhang (2019)	Chronological age	514	Whole blood
LuA2019	DNAmTL	Ake T. Lu (2019)	Leukocyte telomere length	140	Whole blood
epiTOC2	epiTOC2	Andrew E. Teschendorff (2020)	Mitotic divisions	163	Whole blood
ShirebyG2020	Cortical clock	Gemma L Shireby (2020)	Chronological age	347	Brain cortex
DunedinPACE	DunedinPACE	Daniel W Belsky (2022)	Pace of ageing	173	Blood
PCGrimAge	PCGrimAge	Albert T. Higgins-Chen (2022)	GrimAge estimated from DNAm data	78464	Blood
BernabeuE2023c	cAge	Elena Bernabeu (2023)	Chronological age	3225	Blood
LuA2023p2	Pan-mammalian clock2	Ake T. Lu (2023)	Age-to-Lifespan Ratio	816	59 tissues across 185 mammalian species
LuA2023p3	Pan-mammalian clock3	Ake T. Lu (2023)	Age-to-Maturity Ratio	760	59 tissues across 185 mammalian species

1. Citation

If you used this package in your research, please cite us: Wang et al., 2023

@article{Wang2023,
  title={Insights into ageing rates comparison across tissues from recalibrating cerebellum DNA methylation clock},
  author={Wang, Yucheng and Grant, Olivia A and Zhai, Xiaojun and McDonald-Maier, Klaus D and Schalkwyk, Leonard C},
  journal={GeroScience},
  pages={1--18},
  year={2023},
  publisher={Springer}
}

2. Usage

2.1 Installation

Install from Github

## Make sure 'devetools' is installed in your R
# install.packages("devtools")
devtools::install_github("yiluyucheng/dnaMethyAge")

2.2 How to use

Start a R work environment

(1) Predict epigenetic age from DNA methylation data

library('dnaMethyAge')

## prepare betas dataframe
data('subGSE174422') ## load example betas

print(dim(betas)) ## probes in row and samples in column
# 485577 8

availableClock() ## List all supported clocks
# "HannumG2013"  "HorvathS2013" "LevineM2018"  "ZhangQ2019"   "ShirebyG2020"  "YangZ2016"    "ZhangY2017"

clock_name <- 'HorvathS2013'  # Select one of the supported clocks.
## Use Horvath's clock with adjusted-BMIQ normalisation (same as Horvath's paper)
horvath_age <- methyAge(betas, clock=clock_name)

print(horvath_age)
#                         Sample     mAge
# 1 GSM5310260_3999979009_R02C02 74.88139
# 2 GSM5310261_3999979017_R05C01 62.36400
# 3 GSM5310262_3999979018_R02C02 68.04759
# 4 GSM5310263_3999979022_R02C01 61.62691
# 5 GSM5310264_3999979027_R02C01 59.65161
# 6 GSM5310265_3999979028_R01C01 60.95991
# 7 GSM5310266_3999979029_R04C02 52.48954
# 8 GSM5310267_3999979031_R06C02 64.29711

More age models will be added in the future, please get contact if you would like me to add a new clock.

(2) Predict epigenetic age and calculate age acceleration

library('dnaMethyAge')

## prepare betas dataframe
data('subGSE174422') ## load example betas and info

print(dim(betas)) ## probes in row and samples in column
# 485577 8
print(info) ##  info should be a dataframe which includes at least two columns: Sample, Age.
#                         Sample  Age    Sex
# 1 GSM5310260_3999979009_R02C02 68.8 Female
# 2 GSM5310261_3999979017_R05C01 45.6 Female
# 3 GSM5310262_3999979018_R02C02 67.4 Female
# 4 GSM5310263_3999979022_R02C01 45.6 Female
# 5 GSM5310264_3999979027_R02C01 62.5 Female
# 6 GSM5310265_3999979028_R01C01 45.1 Female
# 7 GSM5310266_3999979029_R04C02 53.2 Female
# 8 GSM5310267_3999979031_R06C02 63.8 Female


clock_name <- 'HorvathS2013'  # Select one of the supported clocks, try: availableClock()
## Apply Horvath's clock and calculate age acceleration
## Use Horvath's clock with adjusted-BMIQ normalisation (same as Horvath's paper)
horvath_age <- methyAge(betas, clock=clock_name, age_info=info, fit_method='Linear', do_plot=TRUE)

print(horvath_age)
#                         Sample  Age    Sex     mAge Age_Acceleration
# 1 GSM5310260_3999979009_R02C02 68.8 Female 74.88139         7.334461
# 2 GSM5310261_3999979017_R05C01 45.6 Female 62.36400         3.318402
# 3 GSM5310262_3999979018_R02C02 67.4 Female 68.04759         1.013670
# 4 GSM5310263_3999979022_R02C01 45.6 Female 61.62691         2.581311
# 5 GSM5310264_3999979027_R02C01 62.5 Female 59.65161        -5.586763
# 6 GSM5310265_3999979028_R01C01 45.1 Female 60.95991         2.097534
# 7 GSM5310266_3999979029_R04C02 53.2 Female 52.48954        -9.340977
# 8 GSM5310267_3999979031_R06C02 63.8 Female 64.29711        -1.417638

#### To calculate PCGrimAge, 'age_info' should be a dataframe that contains sample ID, age, sex information ####
PC_GrimAge <- methyAge(betas, clock = "PCGrimAge", age_info=info)

By default, methyAge would plot the age prediction results and the distribution of age acceleration, to save the plot:

pdf('savename.pdf', width=4.3, height=6)
horvath_age <- methyAge(betas, clock=clock_name, age_info=info, fit_method='Linear', do_plot=TRUE)
dev.off()

Here is the result plot(very nice!):

Now you can try other clocks by simply redefine the 'clock_name' and keep other codes unedited. Normally, the clock of Horvath2013 costs the highest amount of time due to its unefficient normalisation steps, for other clocks such as Hannum2013, the overall running time is very short. Please refer to the below code to learn more about how to use the method.

library('dnaMethyAge')

help(methyAge)

Lastly, I provide the epiginetic age prediction results in four clocks for GSE147221 in below(click on the image to have a more clear look).

3. Attention

The original model of Hannum2013 not only uses the 72 CpG sites, but also includes covariates gender, BMI, diabetes status, ethnicity and batch, howevever, the authors did not provide the coefficients for those covariates. In this method, I only used the 72 CpG sites to calculate the Hannum2013 age, and this is the commom practice in this field.

The four clocks' prediciton performance may vary in different datasets, and the Levine2018 also known as PhenoAge was not directly trained on chronological age.

4. Contact me

wangyucheng511@gmail.com