multiFSSEMiPALM2 {fssemR}

multiFSSEMiPALM2

Description

Implementing FSSELM algorithm for network inference. If Xs is identify for different conditions, multiFSSEMiPALM will be use, otherwise, please use multiFSSEMiPALM2 for general cases

Usage

multiFSSEMiPALM2(
  Xs,
  Ys,
  Bs,
  Fs,
  Sk,
  sigma2,
  lambda,
  rho,
  Wl,
  Wf,
  p,
  maxit = 100,
  inert = inert_opt("linear"),
  threshold = 1e-06,
  verbose = TRUE,
  sparse = TRUE,
  trans = FALSE,
  B2norm = NULL,
  strict = FALSE
)

Arguments

Xs

eQTL matrices

Ys

Gene expression matrices

Bs

initialized GRN-matrices

Fs

initialized eQTL effect matrices

Sk

eQTL index of genes

sigma2

initialized noise variance from ridge regression

lambda

Hyperparameter of lasso term in FSSEM

rho

Hyperparameter of fused-lasso term in FSSEM

Wl

weight matrices for adaptive lasso terms

Wf

weight matrix for adaptive fused lasso term

p

number of genes

maxit

maximum iteration number. Default 100

inert

inertial function for iPALM. Default as k-1/k+2

threshold

convergence threshold. Default 1e-6

verbose

Default TRUE

sparse

Sparse Matrix or not

trans

Fs matrix is transposed to k x p or not. If Fs from ridge regression, trans = TRUE, else, trans = FALSE

B2norm

B2norm matrices generated from ridge regression. Default NULL.

strict

Converge strictly or not. Default False

Value

fit List of FSSEM model

Bs

coefficient matrices of gene regulatory networks

Fs

coefficient matrices of eQTL-gene effect

mu

Bias vector

sigma2

estimate of covariance in SEM

Examples

seed = 1234
N = 100                                           # sample size
Ng = 5                                            # gene number
Nk = 5 * 3                                        # eQTL number
Ns = 1                                            # sparse ratio
sigma2 = 0.01                                     # sigma2
set.seed(seed)
library(fssemR)
data = randomFSSEMdata(n = N, p = Ng, k = Nk, sparse = Ns, df = 0.3, sigma2 = sigma2,
                       u = 5, type = "DG", nhub = 1, dag = TRUE)
## If we assume that different condition has different genetics perturbations (eQTLs)
data$Data$X = list(data$Data$X, data$Data$X)
## gamma = cv.multiRegression(data$Data$X, data$Data$Y, data$Data$Sk, ngamma = 20, nfold = 5,
##                            N, Ng, Nk)
gamma = 0.6784248     ## optimal gamma computed by cv.multiRegression
fit   = multiRegression(data$Data$X, data$Data$Y, data$Data$Sk, gamma, N, Ng, Nk,
                      trans = FALSE)
Xs    = data$Data$X
Ys    = data$Data$Y
Sk    = data$Data$Sk


cvfitc <- cv.multiFSSEMiPALM2(Xs = Xs, Ys = Ys, Bs = fit$Bs, Fs = fit$Fs, Sk = Sk,
                             sigma2 = fit$sigma2, nlambda = 5, nrho = 5,
                             nfold = 5, p = Ng, q = Nk, wt = TRUE)

fitc0 <- multiFSSEMiPALM2(Xs = Xs, Ys = Ys, Bs = fit$Bs, Fs = fit$Fs, Sk = Sk,
                          sigma2 = fit$sigma2, lambda = cvfitc$lambda, rho = cvfitc$rho,
                          Wl = inverseB(fit$Bs), Wf = flinvB(fit$Bs),
                          p = Ng, maxit = 100, threshold = 1e-5, sparse = TRUE, 
                          verbose = TRUE, trans = TRUE, strict = TRUE)


(TPR(fitc0$Bs[[1]], data$Vars$B[[1]]) + TPR(fitc0$Bs[[2]], data$Vars$B[[2]])) / 2
(FDR(fitc0$Bs[[1]], data$Vars$B[[1]]) + FDR(fitc0$Bs[[2]], data$Vars$B[[2]])) / 2
TPR(fitc0$Bs[[1]] - fitc0$Bs[[2]], data$Vars$B[[1]] - data$Vars$B[[2]])
FDR(fitc0$Bs[[1]] - fitc0$Bs[[2]], data$Vars$B[[1]] - data$Vars$B[[2]])