meta_analysis_region_based.rmd

---
title: "Single cpg meta analysis in male"
author: "Lanyu Zhang, Lissette Gomez, Tiago C. Silva, Lily Wang"
date: "`r Sys.Date()`"
output:
  rmarkdown::html_document:
    theme: lumen
    highlight: kate
    toc: true
    number_sections: true
    df_print: paged
    code_download: true
    toc_float:
      collapsed: yes
    toc_depth: 3
editor_options:
  chunk_output_type: inline    
---

```{r setup, include = FALSE}
knitr::opts_chunk$set(echo = TRUE)
```

# Data retrival

```{R, message = FALSE, warning = FALSE, results = 'hide'}
library(dplyr)
library(tidyr)
```

```{R}
data.dir <- "../meta_analysis_region_based_results/"
data.gasparoni <- "../DATASETS/GASPARONI/step12_region_pval/"
data.london <- "../DATASETS/LONDON/step11_region_pval/"
data.mtsinai <- "../DATASETS/MtSinai/step11_region_pval/"
data.rosmap <- "../DATASETS/ROSMAP/step12_region_pval/"
data.annot <- "../../coMethDMR_metaAnalysis/DATA_annotations/"
for(p in grep("dir",ls(),value = T)) dir.create(get(p),recursive = TRUE,showWarnings = FALSE)
```

# Prepare and merge datasets for meta analysis

```{R, include = FALSE}
load(file.path(data.gasparoni, "GASPARONI_final_region_based_lm_bacon.rda"))
load(file.path(data.london, "LONDON_final_region_based_lm_bacon.rda"))
load(file.path(data.mtsinai, "MTSINAI_final_region_based_lm_bacon.rda"))
load(file.path(data.rosmap, "ROSMAP_final_region_based_lm_bacon.rda"))
```

```{R}
preMeta <- function(data){
  
  colnames(data)[grep("inputRegion",colnames(data))] <- "inputRegion"
  
  ### select the most sig cometh region for each input region
  pva.col <- grep("_pValue.bacon",colnames(data),value = TRUE)
  
  # !! is used to unquote an input 
  # https://dplyr.tidyverse.org/articles/programming.html
  data_sig <- data %>%
    group_by(inputRegion) %>%
    filter((!!as.symbol(pva.col)) == min(!!as.symbol(pva.col)))
  
  data.frame(data_sig, stringsAsFactors = FALSE)
}
```

## For female

```{R}
gasparoni_female <- preMeta(data = gasparoni_female_bacon_df)
london_female <- preMeta(data = london_female_bacon_df)
mtsinai_female <- preMeta(data = mtsinai_female_bacon_df)
rosmap_female <- preMeta(data = rosmap_female_bacon_df)

cohort_ls <- list(
  GASPARONI = gasparoni_female,
  LONDON = london_female,
  MTSINAI = mtsinai_female,
  ROSMAP = rosmap_female
)

### outer join input region
female_multi_cohorts <- Reduce(
  function(x,y, ...) merge(x, y, by = "inputRegion", all = TRUE, ...),
  cohort_ls
)
# dim(female_multi_cohorts)
```

## For male

```{R}
gasparoni_male <- preMeta(data = gasparoni_male_bacon_df)
london_male <- preMeta(data = london_male_bacon_df)
mtsinai_male <- preMeta(data = mtsinai_male_bacon_df)
rosmap_male <- preMeta(data = rosmap_male_bacon_df)

cohort_ls <- list(
  GASPARONI = gasparoni_male,
  LONDON = london_male,
  MTSINAI = mtsinai_male,
  ROSMAP = rosmap_male
)

### outer join input region
male_multi_cohorts <- Reduce(
  function(x,y, ...) merge(x, y, by = "inputRegion", all = TRUE, ...),
  cohort_ls
)
# dim(male_multi_cohorts)
```

## For stageSex intersection

```{R}
gasparoni_intxn <- preMeta(data = gasparoni_stageSexMale_bacon_df)
london_intxn <- preMeta(data = london_stageSexMale_bacon_df)
mtsinai_intxn <- preMeta(data = mtsinai_stageSexMale_bacon_df)
rosmap_intxn <- preMeta(data = rosmap_stageSexMale_bacon_df)

cohort_ls <- list(
  GASPARONI = gasparoni_intxn,
  LONDON = london_intxn,
  MTSINAI = mtsinai_intxn,
  ROSMAP = rosmap_intxn
)

### outer join input region
intxn_multi_cohorts <- Reduce(
  function(x,y, ...) merge(x, y, by = "inputRegion", all = TRUE, ...),
  cohort_ls
)
# dim(intxn_multi_cohorts)
```

# Meta analysis

```{R, message = FALSE, warning = FALSE, results = 'hide'}
library(dplyr)
library(meta)
library(doParallel)
```

## Create a function for meta analysis

```{R}
metaAnalysis <- function(data,
                         parallel = FALSE,
                         estimate = "Estimate",
                         std = "StdErr"){
  
  meta_df <- plyr::adply(
    .data = data, .margins = 1, .fun =  function(rowOne_df){
      
      est <- rowOne_df[grep(estimate, colnames(rowOne_df))] %>% as.numeric
      
      direction <-  paste(
        ifelse(is.na(est), ".", ifelse(est > 0, "+", "-")), collapse = "")
      
      se <- rowOne_df[grep(std, colnames(rowOne_df))] %>% as.numeric
      
      cohort <- gsub(paste0("_", std),"",grep(std,colnames(rowOne_df),value = T))
      
      rowOne_reform_df <- data.frame(
        cohort = cohort,
        est = est,
        se = se,
        stringsAsFactors = FALSE)
      
      f <- metagen(
        TE = est,
        seTE = se,
        data = rowOne_reform_df)
      
      tibble::tibble(
        inputRegion = rowOne_df$inputRegion,
        estimate = f$TE.fixed,
        se = f$seTE.fixed,
        pVal.fixed = f$pval.fixed,
        pVal.random = f$pval.random,
        pValQ = f$pval.Q,
        direction = direction)},
    
    .progress = "time",
    .parallel = parallel,
    .id = NULL)
  
  ### create final pVal
  meta_df$pVal.final <- ifelse(
    meta_df$pValQ > 0.05, meta_df$pVal.fixed, meta_df$pVal.random)

  ### calculate fdr
  meta_df$fdr <- p.adjust(meta_df$pVal.final, method = "fdr")

  ### order meta_df
  meta_final_df <- meta_df[
    , c(grep("_",colnames(meta_df),invert = T),
        grep("_",colnames(meta_df),invert = F))]

  meta_final_df <- meta_final_df[order(meta_final_df$pVal.final),]
  
  row.names(meta_final_df) <- NULL
  
  meta_final_df
}
```

## For female

```{R, eval = FALSE}
doParallel::registerDoParallel(cores = parallel::detectCores()/2)
female_meta_bacon_df <- metaAnalysis(
  data = female_multi_cohorts,
  parallel = TRUE,
  estimate = "Estimate.bacon",
  std = "StdErr.bacon"
)
```

```{R, include = FALSE, eval = FALSE}
save(
  female_meta_bacon_df,
  file = file.path(data.dir, "region_based_female_meta_bacon_df.rda")
)
```

## For male

```{R, eval = FALSE}
doParallel::registerDoParallel(cores = parallel::detectCores()/2)
male_meta_bacon_df <- metaAnalysis(
  data = male_multi_cohorts,
  parallel = TRUE,
  estimate = "Estimate.bacon",
  std = "StdErr.bacon"
)
```

```{R, include = FALSE, eval = FALSE}
save(
  male_meta_bacon_df,
  file = file.path(data.dir, "region_based_male_meta_bacon_df.rda")
)
```

## For stageSex interaction

```{R, eval = FALSE}
doParallel::registerDoParallel(detectCores()/2)
intxn_meta_bacon_df <- metaAnalysis(
  data = intxn_multi_cohorts,
  parallel = TRUE,
  estimate = "stageSexMale_Estimate.bacon",
  std = "stageSexMale_StdErr.bacon"
)
```

```{R}
save(
  intxn_meta_bacon_df,
  file = file.path(data.dir, "region_based_stageSexMale_meta_bacon_df.rda")
)
```

# StageR analysis

```{R, message = FALSE, warning = FALSE, results = 'hide'}
library(stageR)
```

```{R}
load(file.path(data.dir, "region_based_female_meta_bacon_df.rda"))
load(file.path(data.dir, "region_based_male_meta_bacon_df.rda"))
load(file.path(data.dir, "region_based_stageSexMale_meta_bacon_df.rda"))
```

```{R}
# 1. Set pScreen
pvalueFinal <- merge(
  male_meta_bacon_df[, c("inputRegion", "pVal.final")],
  female_meta_bacon_df[, c("inputRegion", "pVal.final"),],
  by = "inputRegion"
)
colnames(pvalueFinal) <- c("inputRegion", "pvalMale", "pvalFemale")
pScreen <- pmin(pvalueFinal$pvalMale, pvalueFinal$pvalFemale)

# 2. Set pConfirmation
pConfirmation <- merge(
  pvalueFinal,
  intxn_meta_bacon_df[, c("inputRegion", "pVal.final")],
  by = "inputRegion"
)
colnames(pConfirmation)[4] <- "pvalStageSex"
rownames(pConfirmation) <- pConfirmation$inputRegion
pConfirmation <- as.matrix(pConfirmation[,-1])
# identical(rownames(pConfirmation), pvalueFinal$inputRegion)

# 3. stageR analysis
stageRObj <- stageR(
  pScreen = pScreen,
  pConfirmation = pConfirmation,
  pScreenAdjusted = FALSE
)

stageRObj <- stageWiseAdjustment(
  object = stageRObj,
  method = "none",
  alpha = 0.05,
  allowNA = TRUE
)

stage.results <- getAdjustedPValues(
  stageRObj,
  onlySignificantGenes = TRUE,
  order=TRUE
)

# 4. Organize and Merge final results
metaF <- female_meta_bacon_df[, c(1:3,7)]
colnames(metaF) <- c("inputRegion", paste0("metaBaconF_", colnames(metaF)[-1]))

metaM <- male_meta_bacon_df[, c(1:3,7)]
colnames(metaM) <- c("inputRegion", paste0("metaBaconM_", colnames(metaM)[-1]))

metaInt <- intxn_meta_bacon_df[, c(1:3,7)]
colnames(metaInt) <- c("inputRegion", paste0("metaBaconInt_", colnames(metaInt)[-1]))

sexMeta_ls <- list(
  female = metaF,
  male = metaM,
  stageSex = metaInt
)

meta_multi_sex <- Reduce(
  function(x,y, ...) merge(x, y, by = "inputRegion", all = TRUE, ...),
  sexMeta_ls
)

stageMetaResults <- merge(
  stage.results,
  meta_multi_sex,
  by.x = "row.names",
  by.y = "inputRegion"
)
colnames(stageMetaResults)[1] <- "inputRegion"

stageMetaResults <- stageMetaResults %>%
  select(
    inputRegion, padjScreen,
    metaBaconF_estimate, metaBaconF_se, metaBaconF_direction, pvalFemale,
    metaBaconM_estimate, metaBaconM_se, metaBaconM_direction, pvalMale,
    metaBaconInt_estimate, metaBaconInt_se, metaBaconInt_direction, pvalStageSex
  )
```

```{R}
write.csv(
  stageMetaResults,
  file.path(data.dir, "stageR_region_based_bacon_meta_results.csv"),
  row.names = FALSE
)
```

# Add annotation

```{R, message = FALSE, warning = FALSE, results = 'hide'}
library(coMethDMR)
library(doParallel)
```

```{R, include = FALSE}
load(file.path(data.dir, "region_based_female_meta_bacon_df.rda"))
load(file.path(data.dir, "region_based_male_meta_bacon_df.rda"))
load(file.path(data.dir, "region_based_stageSexMale_meta_bacon_df.rda"))
```

## For female

```{R}
female_meta_final_df <- female_meta_bacon_df %>% 
  tidyr::separate(
    col = inputRegion,
    into =  c("chrom", "start", "end"),
    remove = FALSE
  )
female_meta_annot_df <- AnnotateResults(
  female_meta_final_df,
  nCores_int = detectCores()/2
) 

### order vars
orderCol1 <- c(1:4, 57:60, 5:12)
orderCol2 <- grep(
  "GASPARONI|LONDON|MTSINAI|ROSMAP",
  colnames(female_meta_annot_df),
  invert = FALSE)
female_meta_annot_df <- female_meta_annot_df[,c(orderCol1, orderCol2)]
```

```{R, include = FALSE, eval = FALSE}
write.csv(
  female_meta_annot_df,
  file.path(data.dir, "region_based_female_meta_bacon_annot_df.csv"),
  row.names = FALSE
)

write.csv(
  female_meta_annot_df %>% filter(fdr < 0.05),
  file.path(data.dir, "region_based_female_meta_bacon_annot_sig_df.csv"),
  row.names = FALSE
)

```

## For male

```{R}
male_meta_final_df <- male_meta_bacon_df %>% 
  tidyr::separate(
    col = inputRegion,
    into =  c("chrom", "start", "end"),
    remove = FALSE)
male_meta_annot_df <- AnnotateResults(
  male_meta_final_df,
  nCores_int = detectCores()/2) 

### order vars
orderCol1 <- c(1:4, 57:60, 5:12)
orderCol2 <- grep(
  "GASPARONI|LONDON|MTSINAI|ROSMAP",
  colnames(male_meta_annot_df),
  invert = FALSE)
male_meta_annot_df <- male_meta_annot_df[,c(orderCol1, orderCol2)]
```

```{R, include = FALSE, eval = FALSE}
write.csv(
  male_meta_annot_df,
  file.path(data.dir, "region_based_male_meta_bacon_annot_df.csv"),
  row.names = FALSE
)

write.csv(
  male_meta_annot_df %>% filter(fdr < 0.05),
  file.path(data.dir, "region_based_male_meta_bacon_annot_sig_df.csv"),
  row.names = FALSE
)
```

## For stageSex interaction

```{R}
intxn_meta_final_df <- intxn_meta_bacon_df %>% 
  tidyr::separate(
    col = inputRegion,
    into =  c("chrom", "start", "end"),
    remove = FALSE)
intxn_meta_annot_df <- AnnotateResults(
  intxn_meta_final_df,
  nCores_int = detectCores()/2) 

### order vars
orderCol1 <- c(1:4, 53:56, 5:12)
orderCol2 <- grep(
  "GASPARONI|LONDON|MTSINAI|ROSMAP",
  colnames(intxn_meta_annot_df),
  invert = FALSE)
intxn_meta_annot_df <- intxn_meta_annot_df[,c(orderCol1, orderCol2)]
```

```{R, include = FALSE, eval = FALSE}
write.csv(
  intxn_meta_annot_df,
  file.path(data.dir, "region_based_stageSexMale_meta_bacon_annot_df.csv"),
  row.names = FALSE
)
```

## Add GREAT annotation

```{R, message = FALSE, results = "hide"}
library(rGREAT)
library(dplyr)
```

```{R}
addGREAT <- function(dat){
  job_regions <- submitGreatJob(
    gr = dat[, c("chrom", "start", "end")],
    species = "hg19" # or "hg38".
  )
  
  regionsToGenes <- data.frame(plotRegionGeneAssociationGraphs(job_regions))
  regionsToGenes$GREAT_annotation <- ifelse(
    regionsToGenes$distTSS > 0,
    paste0(regionsToGenes$gene, "(+", regionsToGenes$distTSS, ")"),
    paste0(regionsToGenes$gene, "(", regionsToGenes$distTSS, ")")
  )
  
  regionsToGenes <- regionsToGenes[
    ,c("seqnames", "start", "end", "GREAT_annotation")
  ]
  
  great <- group_by(regionsToGenes, seqnames, start, end) %>%
    summarise_each(funs(paste(., collapse = "; ")))
  
  great_df <- as.data.frame(great)
  colnames(great_df)[1] <- "chrom"
  
  merge(dat, great_df, by = c("chrom", "start", "end"))
}
```

```{R}
female_sig <- read.csv(
  file.path(data.dir, "region_based_female_meta_bacon_annot_sig_df.csv")
)
female_sig_great <- addGREAT(female_sig)

male_sig <- read.csv(
  file.path(data.dir, "region_based_male_meta_bacon_annot_sig_df.csv")
)
male_sig_great <- addGREAT(male_sig)
```

```{R}
female_all <- read.csv(
  file.path(data.dir, "region_based_female_meta_bacon_annot_df.csv")
)
female_all_great <- addGREAT(female_all)

male_all <- read.csv(
  file.path(data.dir, "region_based_male_meta_bacon_annot_df.csv")
)
male_all_great <- addGREAT(male_all)
```

## Add chromatin states

```{R, message = FALSE, results = "hide"}
library(GenomicRanges)
```

```{R}
ChmmModels <- readr::read_tsv("https://egg2.wustl.edu/roadmap/data/byFileType/chromhmmSegmentations/ChmmModels/coreMarks/jointModel/final/E073_15_coreMarks_segments.bed",col_names = FALSE)
colnames(ChmmModels) <- c("chr","start","end","state")
states <- readr::read_csv(
  "../../coMethDMR_metaAnalysis/DNAm_RNA/data/chromHMM_labels.csv",
  col_names = FALSE
)
states$X1 <- paste0("E",states$X1)  
ChmmModels$state <- states$X3[match(ChmmModels$state,states$X1)]
ChmmModels.gr <- makeGRangesFromDataFrame(ChmmModels,keep.extra.columns = TRUE)

addState <- function(dat_great){
  dat.gr <- dat_great %>% makeGRangesFromDataFrame()
  hits <- as.data.frame(findOverlaps(dat.gr,ChmmModels.gr))
  
  col.name <- names(values(ChmmModels.gr))[1]
  chrom <- as.data.frame(ChmmModels.gr)
  annot <- data.frame(
    "inputRegion" = dat_great$inputRegion[hits$queryHits],
    "annot" = chrom$state[hits$subjectHits],
    "annot_start" = chrom$start[hits$subjectHits],
    "annot_end" = chrom$end[hits$subjectHits],
    stringsAsFactors = FALSE
  )
  
  annot$annotRegion <- paste0(annot$annot_start, "-", annot$annot_end)
  annot <- annot[, c(1,2,5)]
  
  annotByInputRegion <- group_by(annot, inputRegion) %>%
    summarise_each(funs(paste(., collapse = ";")))
  
  annotByInputRegion <- as.data.frame(annotByInputRegion)
  
  annotByInputRegion <- annotByInputRegion %>%
    select(inputRegion, annot) %>%
    dplyr::rename(state = annot)
  
  merge(dat_great, annotByInputRegion, by = c("inputRegion"))
}
```

```{R}
female_sig_great_state <- addState(female_sig_great)
female_sig_great_state_ordered <- female_sig_great_state[
  , c(1:8, 61:62, 9:60)
]

male_sig_great_state <- addState(male_sig_great)
male_sig_great_state_ordered <- male_sig_great_state[
  , c(1:8, 61:62, 9:60)
]
```

```{R}
write.csv(
  female_sig_great_state_ordered,
  file.path(
    data.dir, 
    "region_based_female_meta_bacon_annot_sig_great_state_df.csv"
  ),
  row.names = FALSE
)

write.csv(
  male_sig_great_state_ordered,
  file.path(
    data.dir, 
    "region_based_male_meta_bacon_annot_sig_great_state_df.csv"
  ),
  row.names = FALSE
)
```

```{R}
female_all_great_state <- addState(female_all_great)
female_all_great_state_ordered <- female_all_great_state[
  , c(1:8, 61:62, 9:60)
]

male_all_great_state <- addState(male_all_great)
male_all_great_state_ordered <- male_all_great_state[
  , c(1:8, 61:62, 9:60)
]
```

```{R}
write.csv(
  female_all_great_state_ordered,
  file.path(
    data.dir, 
    "region_based_female_meta_bacon_annot_great_state_df.csv"
  ),
  row.names = FALSE
)

write.csv(
  male_all_great_state_ordered,
  file.path(
    data.dir, 
    "region_based_male_meta_bacon_annot_great_state_df.csv"
  ),
  row.names = FALSE
)
```

# Session information

```{R}
devtools::session_info()
```