Installation

if (!requireNamespace("BiocManager", quietly = TRUE))
install.packages("BiocManager")
BiocManager::install("RNAAgeCalc")
library(RNAAgeCalc)

Introduction

It has been shown that both DNA methylation and RNA transcription are linked to chronological age and age related diseases. Several estimators have been developed to predict human aging from DNA level and RNA level. Most of the human transcriptional age predictor are based on microarray data and limited to only a few tissues. To date, transcriptional studies on aging using RNASeq data from different human tissues is limited. The aim of this package is to provide a tool for across-tissue and tissue-specific transcriptional age calculation based on Genotype-Tissue Expression (GTEx) RNASeq data (Lonsdale et al. 2013).

Description of RNASeq age calculator

We utilized the GTEx data to construct our across-tissue and tissue-specific transcriptional age calculator. GTEx is a public available genetic database for studying tissue specific gene expression and regulation. GTEx V6 release contains gene expression data at gene, exon, and transcript level of 9,662 samples from 30 different tissues. To avoid the influence of tumor on gene expression, the 102 tumor samples from GTEx V6 release are dropped and the remaining 9,560 samples were used in the subsequent analysis. To facilitate integrated analysis and direct comparison of multiple datasets, we utilized recount2 (Collado-Torres et al. 2017) version of GTEx data, where all samples were processed with the same analytical pipeline. FPKM values were calculated for each individual sample using getRPKM function in Bioconductor package recount.

For the tissue-specific RNASeq age calculator, elastic net (Zou and Hastie 2005) algorithm was used to train the predictors for each individual tissue. Chronological age was response variable whereas logarithm transformed FPKM of genes were predictors. The across-tissue calculator was constructed by first performing differential expression analysis on the RNASeq counts data for each individual tissue. To identify genes consistently differentially expressed across tissues, we adapted the binomial test discussed in de Magalhaes et al. (De Magalhães, Curado, and Church 2009) to find the genes with the largest number of age-related signals. A detailed explanation can be found in our paper.

The package is implemented as follows. For each tissue, signature and sample type (see below for the descriptions), we pre-trained the calculator using elastic net based on the GTEx samples. We saved the pre-trained model coefficients as internal data in the package. The package takes gene expression data as input and then match the input genes to the genes in the internal data. This matching process is automatic so that the users just need to provide gene expression data without having to pull out the internal coefficients.

Usage of RNASeq age calculator

The main functions to calculate RNASeq age are predict_age and predict_age_fromse. Users can use either of them. predict_age function takes data frame as input whereas predict_age_fromse function takes SummarizedExperiment as input. Both functions work in the same way internally. In this section, we explain the arguments in predict_age and predict_age_fromse respectively.

Options in predict_age function

exprdata

exprdata is a matrix or data frame which contains gene expression data with each row represents a gene and each column represents a sample. Users are expected to use the argument exprtype to specify raw count or FPKM (see below). The rownames of exprdata should be gene ids and colnames of exprdata should be sample ids. Here is an example of FPKM expression data:

data(fpkmExample)
head(fpkm)
        SRR2166176 SRR2167642
AFF4     23.478648 21.7803347
ZGLP1     1.880514  2.6874455
CHDH      6.062126  5.9185735
MIR6801   0.000000  0.3882558
ZFP90     6.478456  7.1444598
KDM6B     5.849882  5.1937368

tissue

tissue is a string indicates which tissue the gene expression data is obtained from. Users are expected to provide one of the following tissues. If the tissue argument is not provide or the provided tissue is not in this list, the age predictor trained on all tissues will be used to calculate RNA age.

• blood
• blood_vessel
• brain
• breast
• colon
• esophagus
• heart
• liver
• lung
• muscle
• nerve
• ovary
• pancreas
• pituitary
• prostate
• salivary_gland
• skin
• small_intestine
• spleen
• stomach
• testis
• thyroid
• uterus
• vagina

exprtype

exprtype is either “counts” or “FPKM”. If exprtype is counts, the expression data will be converted to FPKM by count2FPKM automatically and the calculator will be applied on FPKM data. When calculating FPKM, by default gene length is obtained from the package’s internal database. The internal gene length information was obtained from recount2. However, users are able to provide their own gene length information by using genelength argument (see below).

idtype

idtype is a string which indicates the gene id type in exprdata. Default is “SYMBOL”. The following id types are supported.

• SYMBOL
• ENSEMBL
• ENTREZID
• REFSEQ

stype

stype is a string which specifies which version of pre-trained calculators to be used. Two versions are provided. If stype="all", the calculator trained on samples from all races (American Indian/Alaska Native, Asian, Black/African American, and Caucasian) will be used. If stype="Caucasian", the calculator trained on Caucasian samples only will be used. We found that RNA Age signatures could be different in different races (see our paper for details). Thus we provide both the universal calculator and race specific calculator. The race specific calculator for American Indian/Alaska Native, Asian, or Black/African American are not provided due to the small sample size in GTEx data.

signature

signature is a string which indicate the age signature to use when calculating RNA age. This argument is not required.

In the case that this argument is not provided, if tissue argument is also provided and the tissue is in the list above, the tissue specific age signature given by our DESeq2 analysis result on GTEx data will be used. Otherwise, the across tissue signature “GTExAge” will be used.

In the case that this argument is provided, it should be one of the following signatures.

• DESeq2. DESeq2 signature was obtained by performing differential expression analysis on each tissue and select the top differential expressed genes.
• Pearson. Pearson signature represents the genes highly correlated with chronological age by Pearson correlation.
• Dev. Dev signature contains genes with large variation in expression across samples. We adapted the gene selection strategy discussed in (Fleischer et al. 2018), which is a gene must have at least a $$t_1$$-fold difference in expression between any two samples in the training set and at least one sample have expression level > $$t_2$$ FPKM to be included in the prediction models. $$t_1$$ and $$t_2$$ (typically 5 or 10) are thresholds to control the degree of deviance of the genes. We used $$t_1$$ = $$t_2$$ = 10 for most tissues. For some tissues with large sample size, in order to maximize the prediction accuracy while maintaining low computation cost, we increased $$t_1$$ and $$t_2$$ such that the number of genes retained in the model is between 2,000 and 7,000.
• deMagalhaes. deMagalhaes signature contains the 73 age-related genes by (De Magalhães, Curado, and Church 2009).
• GenAge. GenAge signature contains the 307 age-related genes in the Ageing Gene Database (Tacutu et al. 2017).
• GTExAge. GTExAge signature represents the genes consistently differentially expressed across tissues discussed in our paper.
• Peters. Peters signature contains the 1,497 genes differentially expressed with age discussed in (Peters et al. 2015).
• all. “all” represents all the genes used when constructing the RNAAge calculator.

If the genes in exprdata do not cover all the genes in the signature, imputation will be made automatically by the impute.knn function in Bioconductor package impute.

genelength

genelength is a vector of gene length in bp. The size of genelength should be equal to the number of rows in exprdata. This argument is optional. When using exprtype = "FPKM", genelength argument is ignored. When using exprtype = "counts", the raw count will be converted to FPKM. If genelength is provided, the function will convert raw count to FPKM based on the user-supplied gene length. Otherwise, gene length is obtained from the internal database.

chronage

chronage is a data frame which contains the chronological age of each sample. This argument is optional.

If provided, it should be a dataframe with 1st column sample id and 2nd column chronological age. The sample order in chronage doesn’t have to be in the same order as in exprdata. However, the samples in chronage and exprdata should be the same. If some samples’ chronological age are not available, users are expected to set the chronological age in chronage to NA. If chronage contains more than 2 columns, only the first 2 columns will be considered. If more than 30 samples’ chronological age are available, age acceleration residual will be calculated. Age acceleration residual is defined as the residual of linear regression with RNASeq age as dependent variable and chronological age as independent variable.

If this argument is not provided, the age acceleration residual will not be calculated.

Example

chronage = data.frame(sampleid = colnames(fpkm), age = c(30,50))
res = predict_age(exprdata = fpkm, tissue = "brain", exprtype = "FPKM",
chronage = chronage)
signature is not provided, using DESeq2 signature
automatically.
head(res)
             RNAAge ChronAge
SRR2166176 27.32436       30
SRR2167642 49.16285       50

In the above example, we calculated the RNASeq age for 2 samples based on their gene expression data coming from brain. Since the sample size is small, age acceleration residual are not calculated.
Here is an example with sample size > 30:

# This example is just for illustration purpose. It does not represent any
# real data.
# construct a large gene expression data
fpkm_large = cbind(fpkm, fpkm+1, fpkm+2, fpkm+3)
fpkm_large = cbind(fpkm_large, fpkm_large, fpkm_large, fpkm_large)
colnames(fpkm_large) = paste0("sample",1:32)
# construct the samples' chronological age
chronage2 = data.frame(sampleid = colnames(fpkm_large), age = 31:62)
res2 = predict_age(exprdata = fpkm_large, exprtype = "FPKM",
chronage = chronage2)
head(res2)
          RNAAge ChronAge AgeAccelResid
sample1 34.84154       31    -31.254133
sample2 49.77605       32    -16.909233
sample3 65.00385       33     -2.271030
sample4 76.91602       34      9.051533
sample5 82.66541       35     14.211328
sample6 93.07443       36     24.030742

Options in predict_age_fromse function

The main difference between predict_age_fromse and predict_age is that predict_age_fromse takes SummarizedExperiment as input. The se argument is a SummarizedExperiment object.

• The assays(se) should contain gene expression data. The name of assays(se) should be either “FPKM” or “counts”. Use exprtype argument to specify the type of gene expression data provided.
• Users are able to provide the chronological age of samples using colData(se). This is optional. If provided, the column name for chronological age in colData(se) should be “age”. If some samples’ chronological age are not available, users are expected to set the chronological age in colData(se) to NA. If chronological age is not provided, the age acceleration residual will not be calculated.
• Users are able to provide their own gene length using rowData(se). This is also optional. If using exprtype = "FPKM", the provided gene length will be ignored. If provided, the column name for gene length in rowData(se) should be “bp_length”. The function will convert raw count to FPKM by the user-supplied gene length. Otherwise, gene length is obtained from the internal database.
• The other arguments tissue, exprtype, idtype, stype, signature are exactly the same as described in predict_age function.

Example

library(SummarizedExperiment)
colData = data.frame(age = c(40, 50))
se = SummarizedExperiment(assays=list(FPKM=fpkm), colData=colData)
res3 = predict_age_fromse(se = se, exprtype = "FPKM")
head(res3)
             RNAAge ChronAge
SRR2166176 34.84154       40
SRR2167642 49.77605       50

Visualization

We suggest visualizing the results by plotting RNAAge vs chronological age. This can be done by calling makeplot function and passing in the data frame returned by predict_age function.

makeplot(res2)

Session info

sessionInfo()
R version 4.2.0 RC (2022-04-19 r82224)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Ubuntu 20.04.4 LTS

Matrix products: default
BLAS:   /home/biocbuild/bbs-3.15-bioc/R/lib/libRblas.so
LAPACK: /home/biocbuild/bbs-3.15-bioc/R/lib/libRlapack.so

locale:
[1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C
[3] LC_TIME=en_GB              LC_COLLATE=C
[5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8
[7] LC_PAPER=en_US.UTF-8       LC_NAME=C
[11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C

attached base packages:
[1] stats4    stats     graphics  grDevices utils     datasets  methods
[8] base

other attached packages:
[1] SummarizedExperiment_1.26.0 Biobase_2.56.0
[3] GenomicRanges_1.48.0        GenomeInfoDb_1.32.0
[5] IRanges_2.30.0              S4Vectors_0.34.0
[7] BiocGenerics_0.42.0         MatrixGenerics_1.8.0
[9] matrixStats_0.62.0          RNAAgeCalc_1.8.0

loaded via a namespace (and not attached):
[1] backports_1.4.1          Hmisc_4.7-0              BiocFileCache_2.4.0
[4] plyr_1.8.7               splines_4.2.0            BiocParallel_1.30.0
[7] ggplot2_3.3.5            digest_0.6.29            foreach_1.5.2
[10] htmltools_0.5.2          fansi_1.0.3              magrittr_2.0.3
[13] checkmate_2.1.0          memoise_2.0.1            BSgenome_1.64.0
[16] cluster_2.1.3            tzdb_0.3.0               limma_3.52.0
[22] jpeg_0.1-9               colorspace_2.0-3         blob_1.2.3
[25] rappdirs_0.3.3           xfun_0.30                dplyr_1.0.8
[28] crayon_1.5.1             RCurl_1.98-1.6           jsonlite_1.8.0
[31] GEOquery_2.64.0          impute_1.70.0            survival_3.3-1
[34] VariantAnnotation_1.42.0 iterators_1.0.14         glue_1.6.2
[37] gtable_0.3.0             zlibbioc_1.42.0          XVector_0.36.0
[40] DelayedArray_0.22.0      rentrez_1.2.3            scales_1.2.0
[43] DBI_1.1.2                rngtools_1.5.2           derfinder_1.30.0
[46] derfinderHelper_1.30.0   Rcpp_1.0.8.3             progress_1.2.2
[49] htmlTable_2.4.0          bumphunter_1.38.0        foreign_0.8-82
[52] bit_4.0.4                Formula_1.2-4            htmlwidgets_1.5.4
[55] httr_1.4.2               RColorBrewer_1.1-3       ellipsis_0.3.2
[58] pkgconfig_2.0.3          XML_3.99-0.9             farver_2.1.0
[61] nnet_7.3-17              sass_0.4.1               dbplyr_2.1.1
[64] locfit_1.5-9.5           utf8_1.2.2               tidyselect_1.1.2
[67] labeling_0.4.2           rlang_1.0.2              reshape2_1.4.4
[70] AnnotationDbi_1.58.0     munsell_0.5.0            tools_4.2.0
[76] generics_0.1.2           RSQLite_2.2.12           evaluate_0.15
[79] stringr_1.4.0            fastmap_1.1.0            yaml_2.3.5
[82] org.Hs.eg.db_3.15.0      knitr_1.38               bit64_4.0.5
[85] purrr_0.3.4              KEGGREST_1.36.0          nlme_3.1-157
[88] doRNG_1.8.2              xml2_1.3.3               biomaRt_2.52.0
[91] compiler_4.2.0           rstudioapi_0.13          filelock_1.0.2
[94] curl_4.3.2               png_0.1-7                tibble_3.1.6
[97] bslib_0.3.1              stringi_1.7.6            highr_0.9
[100] GenomicFeatures_1.48.0   GenomicFiles_1.32.0      lattice_0.20-45
[103] Matrix_1.4-1             vctrs_0.4.1              pillar_1.7.0
[106] lifecycle_1.0.1          jquerylib_0.1.4          data.table_1.14.2
[109] bitops_1.0-7             rtracklayer_1.56.0       qvalue_2.28.0
[112] R6_2.5.1                 BiocIO_1.6.0             latticeExtra_0.6-29
[115] gridExtra_2.3            codetools_0.2-18         assertthat_0.2.1
[118] rjson_0.2.21             GenomicAlignments_1.32.0 Rsamtools_2.12.0
[121] GenomeInfoDbData_1.2.8   mgcv_1.8-40              parallel_4.2.0
[124] hms_1.1.1                grid_4.2.0               rpart_4.1.16
[127] tidyr_1.2.0              rmarkdown_2.14           base64enc_0.1-3
[130] recount_1.22.0           restfulr_0.0.13         

References

Collado-Torres, Leonardo, Abhinav Nellore, Kai Kammers, Shannon E Ellis, Margaret A Taub, Kasper D Hansen, Andrew E Jaffe, Ben Langmead, and Jeffrey T Leek. 2017. “Reproducible Rna-Seq Analysis Using Recount2.” Nature Biotechnology 35 (4): 319.

De Magalhães, João Pedro, João Curado, and George M Church. 2009. “Meta-Analysis of Age-Related Gene Expression Profiles Identifies Common Signatures of Aging.” Bioinformatics 25 (7): 875–81.

Fleischer, Jason G, Roberta Schulte, Hsiao H Tsai, Swati Tyagi, Arkaitz Ibarra, Maxim N Shokhirev, Ling Huang, Martin W Hetzer, and Saket Navlakha. 2018. “Predicting Age from the Transcriptome of Human Dermal Fibroblasts.” Genome Biology 19 (1): 221.

Lonsdale, John, Jeffrey Thomas, Mike Salvatore, Rebecca Phillips, Edmund Lo, Saboor Shad, Richard Hasz, et al. 2013. “The Genotype-Tissue Expression (Gtex) Project.” Nature Genetics 45 (6): 580.

Peters, Marjolein J, Roby Joehanes, Luke C Pilling, Claudia Schurmann, Karen N Conneely, Joseph Powell, Eva Reinmaa, et al. 2015. “The Transcriptional Landscape of Age in Human Peripheral Blood.” Nature Communications 6: 8570.

Tacutu, Robi, Daniel Thornton, Emily Johnson, Arie Budovsky, Diogo Barardo, Thomas Craig, Eugene Diana, et al. 2017. “Human Ageing Genomic Resources: New and Updated Databases.” Nucleic Acids Research 46 (D1): D1083–D1090.

Zou, Hui, and Trevor Hastie. 2005. “Regularization and Variable Selection via the Elastic Net.” Journal of the Royal Statistical Society: Series B (Statistical Methodology) 67 (2): 301–20.