Exposome Data Integration with Omic Data
19 May 2021
Abstract
This is an introductory guide to integration analysis between exposome and omics data with R package omicRexposome. The document illustrates two types of analysis: 1) Association analysis, that are performed between exposome and a single omic data-set; and 2) Integration analysis where multiple data-sets, including exposome data, are analysed at the same time.Package
omicRexposome 1.14.0
Contents
1 Introduction
omicRexposome is an R package designed to work join with rexposome. The aim of omicRexposome is to perform analysis joining exposome and omic data with the goal to find the relationship between a single or set of exposures (external exposome) and the behavior of a gene, a group of CpGs, the level of a protein, etc. Also to provide a series of tools to analyse exposome and omic data using standard methods from Biocondcutor.
1.1 Installation
omicRexposome is currently in development and not available from CRAN nor Bioconductor. Anyway, the package can be installed by using devtools R package and taking the source from Bioinformatic Research Group in Epidemiology’s GitHub repository.
This can be done by opening an R session and typing the following code:
devtools::install_github("isglobal-brge/omicRexposome")User must take into account that this sentence do not install the packages’ dependencies.
1.2 Pipeline
Two different types of analyses can be done with omicRexposome:
| Analysis | omicRexposome function |
|---|---|
| Association Study | association |
| Integration Study | crossomics |
Both association and integration studies are based in objects of class MultiDataSet. A MultiDataSet object is a contained for multiple layers of sample information. Once the exposome data and the omics data are encapsulated in a MultiDataSet the object can be used for both association and integration studies.
The method association requires a MultiDataSet object having to types of information: the exposome data from an ExposomeSet object and omic information from objects of class ExpressionSet, MethylationSet, SummarizedExperiment or others. ExposomeSet objects are created with functions read_exposome and load_exposome from rexposome R package (see next section Loading Exposome Data) and encapsulates exposome data. The method crossomics expects a MultiDataSet with any number of different data-sets (at last two). Compared with association method, crossomics do not requires an ExposomeSet.
1.3 Exposome and Omic Data
In order to illustrate the capabilities of omicRexposome and the exposome-omic analysis pipeline, we will use the data from BRGdata package. This package includes different omic-sets including methylation, transcriptome and proteome data-sets and an exposome-data set.
2 Analysis
omicRexposome and MultiDataSet R packages are loaded using the standard library command:
library(omicRexposome)
library(MultiDataSet)2.1 Association Studies
The association studies are performed using the method association. This method requires, at last four, augments:
- Argument
objectshould be filled with aMultiDataSetobject. - Argument
formulashould be filled with an expression containing the covariates used to adjust the model. - Argument
expsetshould be filled with the name that the exposome-set receives in theMultiDataSetobject. - Argument
omicsetshould be filled with the name that the omic-set receives in theMultiDataSetobject.
The argument formula should follow the pattern: ~sex+age. The method association will fill the formula placing the exposures in the ExposomeSetm between ~ and the covariates sex+age.
association implements the limma pipeline using lmFit and eBayes in the extraction methods from MultiDataSet. The method takes care of the missing data in exposures, outcomes and omics data and locating and is subsets both data-sets, exposome data and omic data, by common samples. The argument method allows to select the fitting method used in lmFit. By default it takes the value "ls" for least squares but it can also takes "robust" for robust regression.
The following subsections illustrates the usage of association with different types of omics data: methylome, transcriptome and proteome.
2.1.1 Exposome - Transcriptome Data Association
First we get the exposome data from BRGdata package that we will use in the whole section.
data("brge_expo", package = "brgedata")
class(brge_expo)## [1] "ExposomeSet"
## attr(,"package")
## [1] "rexposome"The aim of this analysis is to perform an association test between the gene expression levels and the exposures. So the first point is to obtain the transcriptome data from the brgedata package.
data("brge_gexp", package = "brgedata")The association studies between exposures and transcriptome are done in the same way that the ones with methylome. The method used is association, that takes as input an object of MultiDataSet class with both exposome and expression data.
mds <- createMultiDataSet()
mds <- add_genexp(mds, brge_gexp)
mds <- add_exp(mds, brge_expo)
gexp <- association(mds, formula=~Sex+Age,
expset = "exposures", omicset = "expression")We can have a look to the number of hits and the lambda score of each analysis with the methods tableHits and tableLambda, seen in the previous section.
hit <- tableHits(gexp, th=0.001)
lab <- tableLambda(gexp)
merge(hit, lab, by="exposure")## exposure hits lambda
## 1 BPA_p 19 0.9072377
## 2 BPA_t1 27 0.8807316
## 3 BPA_t3 56 0.9391129
## 4 Ben_p 19 0.8013466
## 5 Ben_t1 12 0.8234104
## 6 Ben_t2 9 0.8393350
## 7 Ben_t3 21 0.8301203
## 8 NO2_p 32 1.0281960
## 9 NO2_t1 16 0.7942881
## 10 NO2_t2 35 1.1482314
## 11 NO2_t3 31 0.8770931
## 12 PCB118 59 0.9308472
## 13 PCB138 38 1.0726221
## 14 PCB153 51 1.1743989
## 15 PCB180 17 0.9790750Since most of all models have a lambda under one, we should consider use Surrogate Variable Analysis. This can be done using the same association method but by setting the argument sva to "fast" so the pipeline of isva and SmartSVA R packages is applied. If sva is set to "slow" the applied. pipeline is the one from sva R package.
gexp <- association(mds, formula=~Sex+Age,
expset = "exposures", omicset = "expression", sva = "fast")We can re-check the results creating the same table than before:
hit <- tableHits(gexp, th=0.001)
lab <- tableLambda(gexp)
merge(hit, lab, by="exposure")## exposure hits lambda
## 1 BPA_p 50 0.9874152
## 2 BPA_t1 51 0.9453795
## 3 BPA_t3 61 0.9842216
## 4 Ben_p 76 1.0117733
## 5 Ben_t1 64 1.0115515
## 6 Ben_t2 71 1.0089834
## 7 Ben_t3 59 0.9969123
## 8 NO2_p 78 1.0117151
## 9 NO2_t1 68 1.0056950
## 10 NO2_t2 69 1.0210000
## 11 NO2_t3 49 0.9802407
## 12 PCB118 129 1.0518170
## 13 PCB138 67 1.0094139
## 14 PCB153 58 0.9924482
## 15 PCB180 67 0.9973381The objects of class ResultSet have a method called plotAssociation that allows to create QQ Plots (that are another useful way to see if there are some inflation/deflation in the P-Values).
gridExtra::grid.arrange(
plotAssociation(gexp, rid="Ben_p", type="qq") +
ggplot2::ggtitle("Transcriptome - Pb Association"),
plotAssociation(gexp, rid="BPA_p", type="qq") +
ggplot2::ggtitle("Transcriptome - THM Association"),
ncol=2
)
Following this line, the same method plotAssociation can be used to create volcano plots.
gridExtra::grid.arrange(
plotAssociation(gexp, rid="Ben_p", type="volcano", tPV=-log10(1e-04)) +
ggplot2::ggtitle("Transcriptome - Pb Association"),
plotAssociation(gexp, rid="BPA_p", type="volcano", tPV=-log10(1e-04)) +
ggplot2::ggtitle("Transcriptome - THM Association"),
ncol=2
)
2.1.2 Exposome - Proteome Data Association
The proteome data-set included in brgedata has 47 proteins for 90 samples.
data("brge_prot", package="brgedata")
brge_prot## ExpressionSet (storageMode: lockedEnvironment)
## assayData: 47 features, 90 samples
## element names: exprs
## protocolData: none
## phenoData
## sampleNames: x0001 x0002 ... x0090 (90 total)
## varLabels: age sex
## varMetadata: labelDescription
## featureData
## featureNames: Adiponectin_ok Alpha1AntitrypsinAAT_ok ...
## VitaminDBindingProte_ok (47 total)
## fvarLabels: chr start end
## fvarMetadata: labelDescription
## experimentData: use 'experimentData(object)'
## Annotation:The association analysis between exposures and proteome is also done using association.
mds <- createMultiDataSet()
mds <- add_eset(mds, brge_prot, dataset.type ="proteome")
mds <- add_exp(mds, brge_expo)
prot <- association(mds, formula=~Sex+Age,
expset = "exposures", omicset = "proteome")The tableHits indicates that no association was found between the 47 proteins and the exposures.
tableHits(prot, th=0.001)## exposure hits
## Ben_p Ben_p 0
## Ben_t1 Ben_t1 0
## Ben_t2 Ben_t2 0
## Ben_t3 Ben_t3 0
## BPA_p BPA_p 0
## BPA_t1 BPA_t1 0
## BPA_t3 BPA_t3 0
## NO2_p NO2_p 0
## NO2_t1 NO2_t1 1
## NO2_t2 NO2_t2 0
## NO2_t3 NO2_t3 0
## PCB118 PCB118 0
## PCB138 PCB138 0
## PCB153 PCB153 0
## PCB180 PCB180 0This is also seen in the Manhattan plot for proteins that can be obtained from plotAssociation.
gridExtra::grid.arrange(
plotAssociation(prot, rid="Ben_p", type="protein") +
ggplot2::ggtitle("Proteome - Cd Association") +
ggplot2::geom_hline(yintercept = 1, color = "LightPink"),
plotAssociation(prot, rid="NO2_p", type="protein") +
ggplot2::ggtitle("Proteome - Cotinine Association") +
ggplot2::geom_hline(yintercept = 1, color = "LightPink"),
ncol=2
)
NOTE: A real Manhattan plot can be draw with plot method for ResultSet objects by setting the argument type to "manhattan".
## used (Mb) gc trigger (Mb) max used (Mb)
## Ncells 7655521 408.9 12466268 665.8 12466268 665.8
## Vcells 20965333 160.0 370941968 2830.1 463670390 3537.62.2 Integration Analysis
omicRexposome allows to study the relation between exposures and omic-features from another perspective, different from the association analyses. The integration analysis can be done, in omicRexposome using multi canonical correlation analysis or using multiple co-inertia analysis. The first methods is implemented in R package PMA (CRAN) and the second in omicade4 R package (Bioconductor). The two methods are encapsulated in the crossomics method.
The differences between association and crossomics are that the first method test association between two complete data-sets, by removing the samples having missing values in any of the involved data-sets, and the second try to find latent relationships between two or more sets.
Hence, we need to explore the missing data in the exposome data-set. This can be done using the methods plotMissings and tableMissings from rexposome R package.
library(rexposome)
plotMissings(brge_expo, set = "exposures")
From the plot we can see that more of the exposures have up to 25% of missing values. Hence the first step in the integration analysis is to avoid missing values. so, we perform a fast imputation on the exposures side:
brge_expo <- imputation(brge_expo)crossomics function expects to obtain the different data-sets in a single labelled-list, in the argument called list. The argument method from crossomics function can be set to mcia (for multiple co-inertia analysis) or to mcca (for multi canonical correlation analysis).
The following code shows how to perform the integration of the exposome and the proteome. The method crossomics request a MultiDataSet object as input, containing the data-set to be integrated.
mds <- createMultiDataSet()
mds <- add_genexp(mds, brge_gexp)
mds <- add_eset(mds, brge_prot, dataset.type = "proteome")
mds <- add_exp(mds, brge_expo)
cr_mcia <- crossomics(mds, method = "mcia", verbose = TRUE)
cr_mcia## Object of class 'ResultSet'
## . created with: crossomics
## . sva:
## . method: mcia ( omicade4 )
## . #results: 1 ( error: 0 )
## . featureData: 3
## . expression: 67528x11
## . proteome: 47x3
## . exposures: 15x12As can be seen, crossomics returns an object of class ResultSet. In the integration process, the different data-sets are subset by common samples. This is done taking advantage of MultiDataSet capabilities.
The same is done when method is set to mcca.
cr_mcca <- crossomics(mds, method = "mcca", permute=c(4, 2))
cr_mccaWe used an extra argument (permute) into the previous call to crossomics using multi canonical correlation analysis. This argument allows to set the internal argument corresponding to permutations and iterations, that are used to tune-up internal parameters.
When a ResultSet is generated using crossomics the methods plotHits, plotLambda and plotAssociation can NOT be used. But the plotIntegration will help us to understand what was done. This method allows to provide the colors to be used on the plots:
colors <- c("green", "blue", "red")
names(colors) <- names(mds)The graphical representation of the results from a multiple co-inertia analysis is a composition of four different plots.
plotIntegration(cr_mcia, colors=colors)
The first plot (first row, first column) is the samples space. It illustrates how the different data-sets are related in terms of intra-sample variability (each data-set has a different color). The second plot (first row, second column) shows the feature space. The features of each set are drawn on the same components so the relation between each data-set can be seen (the features are colored depending of the set were they belong).
The third plot (second row, first column) shows the inertia of each component. The two first plots are drawn on the first and second component. Finally, the fourth plot shows the behavior of the data-sets.
A radar plots is obtained when plotIntegration is used on a ResultSet created though multi canonical correlation analysis.
plotIntegration(cr_mcca, colors=colors)## Warning: ggrepel: 2 unlabeled data points (too many overlaps). Consider
## increasing max.overlaps
This plot shows the features of the three data-sets in the same 2D space.The relation between the features can be understood by proximity. This means that the features that clusters, or that are in the same quadrant are related and goes in a different direction than the features in the other quadrants.
rm(cr_mcia, cr_mcca)Session info
## R version 4.1.0 (2021-05-18)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 20.04.2 LTS
##
## Matrix products: default
## BLAS: /home/biocbuild/bbs-3.13-bioc/R/lib/libRblas.so
## LAPACK: /home/biocbuild/bbs-3.13-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
## [9] LC_ADDRESS=C LC_TELEPHONE=C
## [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C
##
## attached base packages:
## [1] parallel stats graphics grDevices utils datasets methods
## [8] base
##
## other attached packages:
## [1] rexposome_1.14.0 MultiDataSet_1.20.0 omicRexposome_1.14.0
## [4] Biobase_2.52.0 BiocGenerics_0.38.0 BiocStyle_2.20.0
##
## loaded via a namespace (and not attached):
## [1] circlize_0.4.12 backports_1.2.1
## [3] Hmisc_4.5-0 corrplot_0.88
## [5] qqman_0.1.8 gmm_1.6-6
## [7] plyr_1.8.6 splines_4.1.0
## [9] BiocParallel_1.26.0 omicade4_1.32.0
## [11] GenomeInfoDb_1.28.0 fastICA_1.2-2
## [13] ggplot2_3.3.3 made4_1.66.0
## [15] pryr_0.1.4 sva_3.40.0
## [17] digest_0.6.27 foreach_1.5.1
## [19] htmltools_0.5.1.1 magick_2.7.2
## [21] fansi_0.4.2 magrittr_2.0.1
## [23] checkmate_2.0.0 memoise_2.0.0
## [25] cluster_2.1.2 limma_3.48.0
## [27] Biostrings_2.60.0 annotate_1.70.0
## [29] matrixStats_0.58.0 isva_1.9
## [31] sandwich_3.0-1 imputeLCMD_2.0
## [33] prettyunits_1.1.1 jpeg_0.1-8.1
## [35] colorspace_2.0-1 blob_1.2.1
## [37] ggrepel_0.9.1 xfun_0.23
## [39] dplyr_1.0.6 crayon_1.4.1
## [41] RCurl_1.98-1.3 jsonlite_1.7.2
## [43] lme4_1.1-27 genefilter_1.74.0
## [45] impute_1.66.0 zoo_1.8-9
## [47] iterators_1.0.13 survival_3.2-11
## [49] glue_1.4.2 gtable_0.3.0
## [51] zlibbioc_1.38.0 XVector_0.32.0
## [53] DelayedArray_0.18.0 shape_1.4.6
## [55] scales_1.1.1 mvtnorm_1.1-1
## [57] DBI_1.1.1 edgeR_3.34.0
## [59] Rcpp_1.0.6 SmartSVA_0.1.3
## [61] xtable_1.8-4 progress_1.2.2
## [63] htmlTable_2.2.1 clue_0.3-59
## [65] flashClust_1.01-2 foreign_0.8-81
## [67] bit_4.0.4 Formula_1.2-4
## [69] stats4_4.1.0 DT_0.18
## [71] glmnet_4.1-1 htmlwidgets_1.5.3
## [73] httr_1.4.2 gplots_3.1.1
## [75] RColorBrewer_1.1-2 calibrate_1.7.7
## [77] ellipsis_0.3.2 farver_2.1.0
## [79] pkgconfig_2.0.3 XML_3.99-0.6
## [81] nnet_7.3-16 sass_0.4.0
## [83] locfit_1.5-9.4 utf8_1.2.1
## [85] labeling_0.4.2 tidyselect_1.1.1
## [87] rlang_0.4.11 reshape2_1.4.4
## [89] AnnotationDbi_1.54.0 munsell_0.5.0
## [91] tools_4.1.0 cachem_1.0.5
## [93] generics_0.1.0 RSQLite_2.2.7
## [95] ade4_1.7-16 evaluate_0.14
## [97] stringr_1.4.0 fastmap_1.1.0
## [99] yaml_2.2.1 knitr_1.33
## [101] bit64_4.0.5 caTools_1.18.2
## [103] purrr_0.3.4 KEGGREST_1.32.0
## [105] nlme_3.1-152 JADE_2.0-3
## [107] leaps_3.1 PMA_1.2.1
## [109] compiler_4.1.0 rstudioapi_0.13
## [111] png_0.1-7 tibble_3.1.2
## [113] bslib_0.2.5.1 stringi_1.6.2
## [115] highr_0.9 RSpectra_0.16-0
## [117] lattice_0.20-44 Matrix_1.3-3
## [119] nloptr_1.2.2.2 tmvtnorm_1.4-10
## [121] vctrs_0.3.8 norm_1.0-9.5
## [123] pillar_1.6.1 lifecycle_1.0.0
## [125] BiocManager_1.30.15 GlobalOptions_0.1.2
## [127] jquerylib_0.1.4 data.table_1.14.0
## [129] bitops_1.0-7 GenomicRanges_1.44.0
## [131] qvalue_2.24.0 pcaMethods_1.84.0
## [133] R6_2.5.0 latticeExtra_0.6-29
## [135] bookdown_0.22 KernSmooth_2.23-20
## [137] gridExtra_2.3 IRanges_2.26.0
## [139] codetools_0.2-18 boot_1.3-28
## [141] MASS_7.3-54 gtools_3.8.2
## [143] assertthat_0.2.1 SummarizedExperiment_1.22.0
## [145] S4Vectors_0.30.0 GenomeInfoDbData_1.2.6
## [147] mgcv_1.8-35 hms_1.1.0
## [149] grid_4.1.0 rpart_4.1-15
## [151] minqa_1.2.4 rmarkdown_2.8
## [153] MatrixGenerics_1.4.0 lsr_0.5
## [155] scatterplot3d_0.3-41 base64enc_0.1-3
## [157] FactoMineR_2.4