R/SqrtLassoSolver.R

In PriceLab/trena-until-01mar2018: Fit transcriptional regulatory networks using gene expression, priors, machine learning

#------------------------------------------------------------------------------------------------------------------------
#' An S4 class to represent a Square Root LASSO solver
#'
#' @import flare
#' @import BiocParallel
#' @import foreach
#' @import methods
#' 
#' @include Solver.R
#' 
#' @name SqrtLassoSolver-class

.SqrtLassoSolver <- setClass("SqrtLassoSolver",
                             contains="Solver",
                             slots = c(regulatorWeights="numeric",
                                       lambda = "numeric",
                                       nCores = "numeric")
                             )
#----------------------------------------------------------------------------------------------------
#' Create a Solver class object using the Square Root LASSO solver
#'
#' @param mtx.assay An assay matrix of gene expression data
#' @param targetGene A designated target gene that should be part of the mtx.assay data
#' @param candidateRegulators The designated set of transcription factors that could be associated
#' with the target gene
#' @param regulatorWeights A set of weights on the transcription factors
#' (default = rep(1, length(tfs)))
#' @param lambda A tuning parameter that determines the severity of the penalty function imposed
#' on the elastic net regression. If unspecified, lambda will be determined via
#' permutation testing (default = numeric(0)).
#' @param nCores An integer specifying the number of computational cores to devote to this
#' square root LASSO solver. This solver is generally quite slow and is greatly sped up when using
#' multiple cores (default = 4)
#' @param quiet A logical denoting whether or not the solver should print output
#' 
#' @return A Solver class object with Square Root LASSO as the solver
#'
#' @seealso  \code{\link{solve.SqrtLasso}}, \code{\link{getAssayData}}
#'
#' @family Solver class objects
#' 
#' @export
#' 
#' @examples
#' load(system.file(package="trena", "extdata/ampAD.154genes.mef2cTFs.278samples.RData"))
#' target.gene <- "MEF2C"
#' tfs <- setdiff(rownames(mtx.sub), target.gene)
#' sqrt.solver <- SqrtLassoSolver(mtx.sub, target.gene, tfs)

SqrtLassoSolver <- function(mtx.assay=matrix(), targetGene, candidateRegulators,
                            regulatorWeights = rep(1, length(candidateRegulators)),
                            lambda = numeric(0), nCores = 4, quiet=TRUE)
{
    if(any(grepl(targetGene, candidateRegulators)))
        candidateRegulators <- candidateRegulators[-grep(targetGene, candidateRegulators)]
    
    candidateRegulators <- intersect(candidateRegulators, rownames(mtx.assay))
    
    stopifnot(length(candidateRegulators) > 0)    
    
    obj <- .SqrtLassoSolver(Solver(mtx.assay=mtx.assay,                        
                                   quiet=quiet,                        
                                   targetGene=targetGene,                        
                                   candidateRegulators=candidateRegulators),  
                            regulatorWeights=regulatorWeights,                        
                            lambda = lambda,                        
                            nCores = nCores                        
                            )
    
    # Send a warning if there's a row of zeros
    if(!is.na(max(mtx.assay)) & any(rowSums(mtx.assay) == 0))
        warning("One or more gene has zero expression; this may cause problems when using Square Root LASSO. You may want to try 'lasso' or 'ridge' instead.")
    
    obj
    
} # SqrtLassoSolver, the constructor
#----------------------------------------------------------------------------------------------------
#' Show the Square Root Lasso Solver
#' 
#' @rdname show.SqrtLassoSolver
#' @aliases show.SqrtLassoSolver
#'
#' @param object An object of the class SqrtLassoSolver
#'
#' @return A truncated view of the supplied object
#'
#' @examples
#' load(system.file(package="trena", "extdata/ampAD.154genes.mef2cTFs.278samples.RData"))
#' target.gene <- "MEF2C"
#' tfs <- setdiff(rownames(mtx.sub), target.gene)
#' sqrt.solver <- SqrtLassoSolver(mtx.sub, target.gene, tfs)
#' show(sqrt.solver)

setMethod('show', 'SqrtLassoSolver',

          function(object) {
              regulator.count <- length(getRegulators(object))
              if(regulator.count > 10){
                  regulatorString <- paste(getRegulators(object)[1:10], collapse=",")
                  regulatorString <- sprintf("%s...", regulatorString);
              }
              else
                  regulatorString <- paste(getRegulators(object), collapse=",")
              
              msg = sprintf("SqrtLassoSolver with mtx.assay (%d, %d), targetGene %s, %d candidate regulators %s,  with %d cores",
                            nrow(getAssayData(object)), ncol(getAssayData(object)),
                            getTarget(object), regulator.count, regulatorString, object@nCores)
              cat (msg, '\n', sep='')
          })
#----------------------------------------------------------------------------------------------------
#' Run the Square Root LASSO Solver
#'
#' @rdname solve.SqrtLasso
#' @aliases run.SqrtLassoSolver solve.SqrtLasso
#' 
#' @description Given SqrtLassoSolver object, use the \code{\link{slim}} function to
#' estimate coefficients for each transcription factor as a predictor of the
#' target gene's expression level.
#' 
#' @param obj An object of class Solver with "sqrtlasso" as the solver string
#'
#' @return A data frame containing the coefficients relating the target gene to
#' each transcription factor, plus other fit parameters.
#'
#' @seealso \code{\link{slim}}, \code{\link{SqrtLassoSolver}}
#'
#' @family solver methods
#'
#' @examples
#' # Load included Alzheimer's data, create a TReNA object with Square Root LASSO as solver,
#' # and run using a few predictors
#'
#' \dontrun{
#' load(system.file(package="trena", "extdata/ampAD.154genes.mef2cTFs.278samples.RData"))
#' target.gene <- "MEF2C"
#'
#' # Designate just 5 predictors and run the solver
#' tfs <- setdiff(rownames(mtx.sub), target.gene)[1:5]
#' sqrt.solver <- SqrtLassoSolver(mtx.sub, target.gene, tfs)
#' tbl <- run(sqrt.solver)
#' }

setMethod("run", "SqrtLassoSolver",

          function (obj){
              
              mtx <- getAssayData(obj)
              target.gene <- getTarget(obj)
              tfs <- getRegulators(obj)
              lambda <- obj@lambda
              nCores <- obj@nCores            
              
              # we don't try to handle tf self-regulation              
              deleters <- grep(target.gene, tfs)              
              if(length(deleters) > 0){                  
                  tfs <- tfs[-deleters]                  
                  if(!obj@quiet)
                      message(sprintf("SqrtLassoSolver removing target.gene from candidate regulators: %s", target.gene))
              }
              
              if( length(tfs) == 0 ) return( data.frame() )
              
              stopifnot(target.gene %in% rownames(mtx))             
              stopifnot(all(tfs %in% rownames(mtx)))              
              stopifnot(class(lambda) %in% c("NULL","numeric"))              
              features <- t(mtx[tfs,,drop=FALSE ])              
              target <- as.numeric(mtx[target.gene,])
              
              if( length(tfs) == 1 ) {                  
                  fit = stats::lm( target ~ features )                  
                  mtx.beta = stats::coef(fit)                  
                  mtx.beta = data.frame( beta = mtx.beta[2] , intercept = mtx.beta[1] )
                  rownames(mtx.beta) = tfs                  
                  return( mtx.beta )                  
              }
              
              # If no lambda, run a binary search for the best lasso using permutation of the data set
              if(length(lambda) == 0){
                  
                  target.mixed <- sample(target)
                  threshold <- 1E-15
                  lambda.change <- 10^(-4)
                  lambda <- 1
                  
                  # Register a BiocParallel instance based on platform
                  if(Sys.info()['sysname'] == "Windows"){
                      BiocParallel::register(BiocParallel::SnowParam(workers = nCores,
                                                                     stop.on.error = FALSE,
                                                                     log = FALSE),
                                             default = TRUE)                      
                  } else{                  
                      BiocParallel::register(BiocParallel::MulticoreParam(workers = nCores,
                                                                          stop.on.error = FALSE,
                                                                          log = FALSE),                                             
                                             default = TRUE)}                  
                  
                  lambda.list <- BiocParallel::bplapply(rep(lambda,30), function(lambda){
                      
                      # Do a binary search
                      step.size <- lambda/2 # Start at 0.5
                      while(step.size > lambda.change){
                          # Get the fit
                          fit <- flare::slim(features, target.mixed, method = "lq", verbose = FALSE, lambda = lambda)
                          # Case 1: nonsense, need to lower lambda
                          if(max(fit$beta) < threshold){
                              lambda <- lambda - step.size
                          }
                          # Case 2: sense, need to raise lambda
                          else{
                              lambda <- lambda + step.size
                          }
                          # Halve the step size and re-scramble the target
                          step.size <- step.size/2
                          target.mixed <- sample(target)
                      }
                      lambda
                  })
                  
                  # Could potentially stop the cluster here
                  
                  # Grab the lambdas and average them
                  lambda.list <- unlist(lambda.list)                                                   
                  lambda <- mean(lambda.list) + (stats::sd(lambda.list)/sqrt(length(lambda.list)))                                 
              }              
              
              # Run square root lasso and return an object of class "slim"              
              fit <- flare::slim(features, target, method = "lq", lambda = lambda, verbose=FALSE)
              
              # Pull out the coefficients        
              mtx.beta <- as.matrix(fit$beta)
              colnames(mtx.beta) <- "beta"
              rownames(mtx.beta) <- colnames(features)
              deleters <- as.integer(which(mtx.beta[,1] == 0))
              if( all( mtx.beta[,1] == 0 ) ) return( data.frame() )
              if(length(deleters) > 0)
                  mtx.beta <- mtx.beta[-deleters, , drop=FALSE]
              
              # put the intercept, admittedly with much redundancy, into its own column
              mtx.beta <- cbind(mtx.beta, intercept=rep(fit$intercept, nrow(mtx.beta)))
              
              mtx.beta <- as.data.frame(mtx.beta)
              
              if( nrow(mtx.beta) > 1 ) {
                  ordered.indices <- order(abs(mtx.beta[, "beta"]), decreasing=TRUE)
                  mtx.beta <- mtx.beta[ordered.indices,]
              }
              
              return(mtx.beta)
          })
#----------------------------------------------------------------------------------------------------