multiFSSEMiPALM: multiFSSEMiPALM

In fssemR: Fused Sparse Structural Equation Models to Jointly Infer Gene Regulatory Network

multiFSSEMiPALM

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

multiFSSEMiPALM(
  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)
## 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.multiFSSEMiPALM(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 <- multiFSSEMiPALM(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]])

fssemR documentation built on March 18, 2022, 7:24 p.m.