R/q_mv2.R

In shimlab/riboHMM2: Analysis tool for Ribo-seq

#' q_mv2
#'
#' \code{q_mv2} is the emission part of the Q function in E-step,
#' i.e.: Sum_{i,t,k} log[fk(x_t)] * L^i_k(t). (with negative sign).
#' \code{q_mv2} uses mean-parametrization to calculates the target function for each k
#' instead of all k.
#'
#' @export
#' @param pars a vector of length 2. c(v[k], m[k])
#' @param X    a list of vectors of observed states x
#' @param E    a vector of normalizing constant for each observed chain in X
#' @param L    a list of matrix L from \code{computeL}
#' @param k    a scalar indicating which state is calculated
#' @return     A scalar, the (negative) value of the target function
#'                that would later be minimized.
#'
#' @examples
#' df <- uORF[1:10]
#' X <- L <- list()
#' E <- c()
#' for (i in 1:length(df)){
#'   X[[i]] <- df[[i]]$x
#'   RNA[[i]] <- df[[i]]$RNA
#'   E[i]=df[[i]]$E;   trans=df[[i]]$trans;
#'   a=df[[i]]$v;      b=df[[i]]$v/df[[i]]$m
#'   la <- forwardAlg(X[[i]], RNA[[i]], trans, a, b, E[i])
#'   lb <- backwardAlg(X[[i]], RNA[[i]], trans, a, b, E[i])
#'   L[[i]] <- computeL(la, lb)
#' }
#' pars <- c(df[[1]]$v, df[[1]]$m)
#'
#' qe <- 0
#' for (k in 1:21){
#'   qe <- qe + q_mv2(pars[c(k,21+k)],X,E,L,k)
#' }
#' print(qe)
#' # qemiss() uses a,b instead of v,m
#' pars <- c(df[[1]]$v, df[[1]]$v/df[[1]]$m)
#' print(qemiss(pars,X,E,L))

q_mv2 <- function(pars, X, E, L, k){
  pars <- abs(pars)
  v <- pars[1]
  m <- pars[2]
  qe = 0
  for (i in 1:length(X)){
    for (t in 1:length(X[[i]])){
      lf_tk <- lnb(X[[i]][t], v, v/m, E[i])
      qe <- qe + lf_tk * L[[i]][t,k]
    }
  }
  return(-qe)
}