--- title: "Sample Metadata Inference" date: "`r BiocStyle::doc_date()`" package: sesame output: rmarkdown::html_vignette fig_width: 6 fig_height: 5 vignette: > %\VignetteEngine{knitr::rmarkdown} %\VignetteIndexEntry{"4. Data Inference"} %\VignetteEncoding{UTF-8} --- SeSAMe implements inference of sex, age, ethnicity. These are valuable information for checking the integrity of the experiment and detecting sample swaps. ```{r inf1, echo=FALSE, message=FALSE} library(sesame) sesameDataCache() sdf = sesameDataGet('EPIC.1.SigDF') ``` # Sex, XCI Sex is inferred based on our curated X-linked probes and Y chromosome probes excluding pseudo-autosomal regions and XCI escapes. Human: ```{r inf2, message=FALSE} sdf = sesameDataGet('EPIC.1.SigDF') inferSex(sdf) inferSexKaryotypes(sdf) ``` Mouse: ```{r nh16, message=FALSE} sdf = sesameDataGet("MM285.1.SigDF") inferSex(sdf) ``` # Ethnicity Ethnicity is inferred using a random forest model trained based on both the built-in SNPs (`rs` probes) and channel-switching Type-I probes. ```{r inf3} sdf = sesameDataGet('EPIC.1.SigDF') inferEthnicity(sdf) ``` # Age & Epigenetic Clock SeSAMe provides age regression through multiple previously established models, e.g., the well-known Horvath 353 model ([Horvath 2013](https://pubmed.ncbi.nlm.nih.gov/24138928/)) which returns the chronological age in the number of years. Here is an example: ```{r inf4, eval=FALSE} betas <- sesameDataGet('HM450.1.TCGA.PAAD')$betas model <- sesameAnno_get("Anno/HM450/Clock_Horvath353.rds") predictAge(betas, model) ``` And MM285 mouse array data using a set of 347 CpGs (see [Zhou et al. 2022](https://www.cell.com/cell-genomics/fulltext/S2666-979X(22)00077-5)) The function returns the age in the number of months. We recommend using SeSAMe preprocessed data as input to the function. Here’s an example: ```{r inf18, message=FALSE, eval=FALSE} library(SummarizedExperiment) betas <- assay(sesameDataGet("MM285.10.SE.tissue"))[,1] model <- sesameAnno_get("Anno/MM285/Clock_Zhou347.rds") predictAge(betas, model) ``` This indicates that this mouse is approximately 1.41 months old. The function looks for overlapping probes and estimates age using the corresponding clock models. Other available epigenetic clocks are ```{r inf20, echo=FALSE, result="asis"} library(knitr) df <- data.frame(rbind( c("Anno/HM450/Clock_Horvath353.rds", 353, "HM450/EPIC", "Horvath 2013 (24138928)"), c("Anno/HM450/Clock_Hannum.rds", 71, "HM450", "Hannum 2013 (23177740)"), c("Anno/HM450/Clock_SkinBlood.rds", 391, "HM450/EPIC", "Horath 2018 (30048243)"), c("Anno/EPIC/Clock_PhenoAge.rds", 514, "HM450/EPIC", "Levine 2018 (29676998)"), c("Anno/MM285/Clock_Zhou347.rds", 347, "MM285", "Zhou 2022") )) colnames(df) <- c("RDS Key", "Platform", "N", "Reference (PMID)") kable(df, caption="Available Epigenetic Clocks") ``` # Copy Number See [Supplemental Vignette](https://zhou-lab.github.io/sesame/v1.16/supplemental.html#cnv) # Cell Count Deconvolution SeSAMe estimates leukocyte fraction using a two-component model.This function works for samples whose targeted cell-of-origin is not related to white blood cells. ```{r inf7, message=FALSE} betas.tissue <- sesameDataGet('HM450.1.TCGA.PAAD')$betas estimateLeukocyte(betas.tissue) ``` # Genomic Privacy The goal of data sanitization is to modifiy IDAT files in place, so they can be released to public domain without privacy leak. This will be achieved by deIdentification. ```{r inf8, message=FALSE, warning=FALSE, include=FALSE} library(sesame) sesameDataCacheAll() ``` Let's take DNA methylation data from the HM450 platform for example. ```{r inf9, eval=FALSE} tmp = tempdir() res_grn = sesameAnno_download("Test/3999492009_R01C01_Grn.idat", dest_dir=tmp) res_red = sesameAnno_download("Test/3999492009_R01C01_Red.idat", dest_dir=tmp) ``` ## De-identify by Masking This first method of deIdentification masks SNP probe intensity mean by zero. As a consequence, the allele frequency will be 0.5. ```{r inf10, eval=FALSE} deIdentify(res_grn$dest_file, sprintf("%s/deidentified_Grn.idat", tmp)) deIdentify(res_red$dest_file, sprintf("%s/deidentified_Red.idat", tmp)) betas1 = getBetas(readIDATpair(sprintf("%s/Test/3999492009_R01C01", tmp))) betas2 = getBetas(readIDATpair(sprintf("%s/deidentified", tmp))) head(betas1[grep('rs',names(betas1))]) head(betas2[grep('rs',names(betas2))]) ``` Note that before deIdentify, the rs values will all be different. After deIdentify, the rs values will all be masked at an intensity of 0.5. ## De-identify by Scrambling This second method of deIdentification will scramble the intensities using a secret key to help formalize a random number. Therefore, randomize needs to be set to TRUE. ```{r inf11, eval=FALSE} my_secret <- 13412084 set.seed(my_secret) deIdentify(res_grn$dest_file, sprintf("%s/deidentified_Grn.idat", tmp), randomize=TRUE) my_secret <- 13412084 set.seed(my_secret) deIdentify(res_red$dest_file, sprintf("%s/deidentified_Red.idat", tmp), randomize=TRUE) betas1 = getBetas(readIDATpair(sprintf("%s/Test/3999492009_R01C01", tmp))) betas2 = getBetas(readIDATpair(sprintf("%s/deidentified", tmp))) head(betas1[grep('rs',names(betas1))]) head(betas2[grep('rs',names(betas2))]) ``` Note that the rs values are scrambled after deIdentify. ## Re-identify To restore order of the deIdentified intensities, one can re-identify IDATs. The reIdentify function can thus restore the scrambled SNP intensities. ```{r inf12, eval=FALSE} my_secret <- 13412084 set.seed(my_secret) reIdentify(sprintf("%s/deidentified_Grn.idat", tmp), sprintf("%s/reidentified_Grn.idat", tmp)) my_secret <- 13412084 set.seed(my_secret) reIdentify(sprintf("%s/deidentified_Red.idat", tmp), sprintf("%s/reidentified_Red.idat", tmp)) betas1 = getBetas(readIDATpair(sprintf("%s/Test/3999492009_R01C01", tmp))) betas2 = getBetas(readIDATpair(sprintf("%s/reidentified", tmp))) head(betas1[grep('rs',names(betas1))]) head(betas2[grep('rs',names(betas2))]) ``` Note that reIdentify restored the values. Subsequently, they are the same as betas1. # Session Info ```{r} sessionInfo() ```