README.md

In vanduttran/dsMO: Client Side Functions for Federated Multi-Omics Analysis on the Virtual Cohort

dsMO: non-disclosive federated Multi-Omics analysis

Introduction

The dsMO package suit, providing a number of non-disclosive federated Multi-Omics analyses, is comprised of three packages:

1. dsMOprimal to install on remote servers with data contributing to the virtual cohort

2. dsMOdual, to install on a remote server different from the previous ones

3. dsMO, to install on the analyst side

Installation

devtools::install_github('sib-swiss/dsSwissKnifeClient')
devtools::install_github('vanduttran/dsMO')

Examples

library(dsMO)

Login data

logindata <- read.table('logindata_BEAt.txt', header=T)

logindata contains login credentials, which can be used to execute as it is. Otherwise, to prompt for login credentials:

logindata <- logindata[, -c(grep('user|password', colnames(logindata)))]

Function for data preprocessing: create raw data matrices

One data block: for federated PCA

dataProc.SingleOmics <- function(conns, symbol) {
    ## dsSwissKnifeClient is required for some manipulation on remote data
    require(dsSwissKnifeClient)

    ## create a data frame tmp (samples x variables) with 5 variables (in columns) from remote CDISC-formatted data
    lapply(conns, function(conn) {
        DSI::datashield.assign(conn, 'tmp', paste0(conn@name, '.cnsim'), 
                               variables=c('LAB_TSC', 'LAB_TRIG', 'LAB_HDL', 'LAB_GLUC_ADJUSTED', 'PM_BMI_CONTINUOUS'), async=T)
    })

    ## filter out samples with missing values and LAB_TSC measure less than 4 from tmp
    dssSubset(symbol='tmp', what='tmp', row.filter='complete.cases(tmp)', datasources=conns)
    dssSubset(symbol='tmp', what='tmp', row.filter='!is.na(LAB_TSC) & as.numeric(as.character(LAB_TSC)) < 4', datasources=conns)

    ## assign customized rownames to tmp then assign the data block tmp to symbol
    rns <- datashield.aggregate(conns, as.symbol('rowNames(tmp)'), async=T)
    lapply(names(conns), function(opn) {
        datashield.assign(conns[opn], symbol, 
                          as.call(list(as.symbol("setRowNames"),
                                       as.symbol("tmp"),
                                       dsSwissKnifeClient:::.encode.arg(rns[[opn]]))),
                          async=T)
    })
}

Two data blocks: for federated RCCA, ComDim, SNF

dataProc.BiOmics <- function(conns, symbol) {
    ## dsSwissKnifeClient is required for some manipulation on remote data
    require(dsSwissKnifeClient)

    ## create a data frame tmp (samples x variables) with 5 variables (in columns) from remote CDISC-formatted data
    lapply(conns, function(conn) {
        DSI::datashield.assign(conn, 'tmp', paste0(conn@name, '.cnsim'), 
                               variables=c('LAB_TSC', 'LAB_TRIG', 'LAB_HDL', 'LAB_GLUC_ADJUSTED', 'PM_BMI_CONTINUOUS'), async=T)
    })

    ## filter out samples with missing values and LAB_TSC measure less than 4 from tmp
    dssSubset(symbol='tmp', what='tmp', row.filter='complete.cases(tmp)', datasources=conns)
    dssSubset(symbol='tmp', what='tmp', row.filter='!is.na(LAB_TSC) & as.numeric(as.character(LAB_TSC)) < 4', datasources=conns)

    ## assign customized rownames to tmp
    rns <- datashield.aggregate(conns, as.symbol('rowNames(tmp)'), async=T)
    lapply(names(conns), function(opn) {
        datashield.assign(conns[opn], 'tmp', 
                          as.call(list(as.symbol("setRowNames"),
                                       as.symbol("tmp"),
                                       dsSwissKnifeClient:::.encode.arg(rns[[opn]]))),
                          async=T)
    })

    ## create two data blocks and assign to symbol
    dssSubset(symbol=symbol[1], what='tmp', col.filter="c('LAB_TSC', 'LAB_TRIG', 'LAB_HDL')", datasources=conns)
    dssSubset(symbol=symbol[2], what='tmp', col.filter="c('LAB_GLUC_ADJUSTED', 'PM_BMI_CONTINUOUS')", datasources=conns)
}

Run exec on different analyses

PCA (Single data block) for the virtual cohort combining those from servers in logins

federatePCA needs to be called at the server with dsMOprimal installed, in loginFD

res.pca <- exec('federatePCA', loginFD=logindata[1,], logins=logindata[1:2,],
                func=dataProc.SingleOmics, symbol='rawDataX')

RCCA (Two data blocks) for the virtual cohort combining those from servers in logins

federateRCCA needs to be called at the server with dsMOprimal installed, in loginFD

res.rcca <- exec('federateRCCA', loginFD=logindata[1,], logins=logindata[1:2,],
                 func=dataProc.BiOmics, symbol=c('rawDataX', 'rawDataY'),
                 0.001, 0.001, TRUE, 
                 dsMO:::.encode.arg(list(nfold = 2,
                                         grid1 = seq(0.001, 1, length = 1),
                                         grid2 = seq(0.001, 1, length = 1))))

ComDim (Multiple data blocks) for the virtual cohort combining those from servers in logins

federateComDim needs to be called at the server with dsMOdual installed, in loginFD

res.comdim <- exec('federateComDim', loginFD=logindata[3,], logins=logindata[1:2,],
                   func=dataProc.BiOmics, symbol=c('rawDataX', 'rawDataY'),
                   2, 'none', 'uniform')

SNF (Multiple data blocks) for the virtual cohort combining those from servers in logins

federateSNF needs to be called at the server with dsMOdual installed, in loginFD

res.snf <- exec('federateSNF', loginFD=logindata[3,], logins=logindata[1:2,],
                func=dataProc.BiOmics, symbol=c('rawDataX', 'rawDataY'))

vanduttran/dsMO documentation built on Oct. 17, 2024, 1:14 a.m.