ZIPPCApn: ZIPPCApn

In YanyZeng/mbDenoise: Zero-Inflated Probabilistic PCA with different distributions, including Poisson, negative-binomial,and logistical normal multinomial (ZIPPCA-Poi,ZIPPCA-NB,ZIPPCA-LNM)

Description

Microbiome data denoising framework (mbDenoise) with zero-inflated probabilistic PCA with Poisson (ZIPPCA-Poi) and negative-binomial model (ZIPPCA-NB), which can be used for downstream statistical analysis including ordination, compositional normalization, differential abundance analysis, etc. mbDenoise with ZIPPCA-NB model is recommended for empirical data analysis.

Usage

ZIPPCApn(
  X,
  V = NULL,
  family = "negative.binomial",
  n.factors = 2,
  rank = FALSE,
  trace = FALSE,
  maxit = 100,
  parallel = TRUE
)

Arguments

X	matrix of observations.
V	vector of the sample covariate.
family	distribution of models. Two options are "poisson" and "negative.binomial". Defaults to "negative.binomial".
n.factors	the rank or number of factors, after dimensional reduction. Defaults to 2.
rank	logical, if TRUE, the rank or number of factors, is chosen from 1 to 5 by HIC (hybrid information criterion). Defaults to FALSE.
trace	logical, defaults to FALSE. if TRUE each current iteration step information will be printed.
maxit	maximum number of iterations within optim and constrOptim function, defaults to 100.
parallel	logical, if TRUE, use parallel toolbox to accelerate.

Value

VLB	variational lower bound of log likelihood
lvs	list of latent variables pi the probabilities of excess zeros factor_scores coordinates or factor scores in low-dimensional subspace factor_scores2 coordinates or factor scores in low-dimensional subspace with defalt rank 2, which is suitable for visualization.
params	list of model parameters factor_coefs_j coefficients of latent variables fator scores or factor loadings factor_coefs_0 taxon-specific intercepts alpha sample-specifc coeffcient that adjusts for the sequencing depth dispersion taxon-specific over-dispersion parameter for negative binomial distribution gamma coeffcients of sample covariate tuo taxon-specific parameter of zero-inflation probability c sample-specific parameter of zero-inflation probability
Q	the underlying composition of microbiome data
muz	the denoised counts of microbiome data
hic	the number of the rank selection, chosen by HIC type information criterion

Examples

n.n = 60
n.w = 100
n.factors = 2
set.seed(1)
si <- diag(n.factors)
me <- c(0,0)
f <- matrix(0,nrow = n.n, ncol = n.factors)
for(i in 1:n.n){
 f[i,] <- rnorm(n.factors, mean = 0, sd = 1)
}
betaj <- matrix(0,nrow = n.w, ncol = n.factors)
for(j in 1:n.w){
  betaj[j,] <- runif(n.factors,-3,3)
}
alpha <- runif(n.n,-5,5)
beta0 <- rep(0,n.w)
g <- rep(0,n.w*0.5*0.5)
gamma <- c(g,-g,rep(0,(n.w-n.w*0.5)))
X_cov<- c(rep(1,n.n/2),rep(0,n.n/2))
ll <- f %*% t(betaj)  +matrix(alpha,n.n,n.w)+matrix(beta0,n.n,n.w,byrow=TRUE)
exp_mat <- exp(ll)
eta_mat <- matrix(0.25,n.n,n.w,byrow=TRUE)
z <- matrix(0,n.n,n.w,byrow = TRUE)
for(i in 1:n.n){
  z[i,] <- rbinom(n.w, size=1, prob=eta_mat[i,])
}
sum <- rowSums((1-z)*exp_mat)
Qn_z <- (1-z)*exp_mat/sum
sum <- rowSums(exp_mat)
Qn <- exp_mat/sum
X <- matrix(0,n.n,n.w,byrow = TRUE)
 for(i in 1:n.n){
  for(j in 1:n.w){
    X[i,j] <- rnbinom(n=1,size=10,mu=exp_mat[i,j])
 }
 }
X[z==1]=0
zerorow <- which(rowSums(X)==0)
if(length(zerorow) >0 ){
   X <- X[-zerorow,];X_cov<-X_cov[-zerorow];f <- f[-zerorow,];
   Qn <- Qn[-zerorow,];Qn_z <- Qn_z[-zerorow,];
}
zerocol <- which(colSums(X)==0)
if(length(zerocol) >0 ){
  X <- X[,-zerocol];betaj <- t(t(betaj)[,-zerocol]);
  Qn <- Qn[,-zerocol];Qn_z <- Qn_z[,-zerocol];
}
re_zinb_cov <- ZIPPCApn(X,X_cov)
re_zinb <- ZIPPCApn(X)

YanyZeng/mbDenoise documentation built on Dec. 18, 2021, 7:25 p.m.