NIFA: Main NIFA function

In wgmao/NIFA: Non-negative Independent Factor Analysis for single cell RNA-seq

Main NIFA function

Description

Main NIFA function

Usage

NIFA(X, K = 6, M = 4, init="sd", S.init=NULL, A.init=NULL, L1.sd = NULL, L2.sd = NULL , verbose=F, beta_threshold = 2e-5, 
     S_threshold = 6e-5, max.iter = 1000, ref=NULL, rho_S_diff=1.5,phi_S_prior=1, a_S_prior=1,b_S_prior=1, a_noise_prior=1, 
     b_noise_prior=1,lambda_A_prior=1, eta_A_prior=0, ELBO=F, A.immediate =F, beta_expect_flag = NULL)

Arguments

X	the data to be processed with genes in rows and samples in columns.
K	number of latent factors (default: 6)
M	maximum number of mixtures associated with each latent factor (default: 4)
init	initialization method and there are three options: "random", "ICA" and "sd" (default: "sd"-simple decomposition)
S.init	initialization values for latent factors S (default: NULL)
A.init	initialization values for loadings A (default: NULL)
L1.sd	regularization parameter L1 in simple decomposition (default: NULL)
L2.sd	regularization parameter L2 in simple decomposition (default: NULL)
verbose	print out intermediate messages (default: F)
beta_threshold	threshold for relateive change of beta (default: 2e-5)
S_threshold	threshold for relative change of S (default: 6e-5)
max.iter	maximum number of iterations (default: 1000)
ref	independent measurement that can be associated with S, e.g., one_hot_encode(celltype) (default: NULL)
rho_S_diff	hyperparameter to initialize rho (default: 1.5)
phi_S_prior	hyperparameter to initialize phi (default: 1)
a_S_prior	prior parameter for S (default: 1)
b_S_prior	prior parameter for S (default: 1)
a_noise_prior	prior paramter for beta (default: 1)
b_noise_prior	prior parameter for beta (default: 1, recommended: prod(dim(X))*5)
lambda_A_prior	prior parameter for A (default: 1)
eta_A_prior	prior parameter for A (default: 0)
ELBO	whether calculate ELBO or not (default: F)
A.immediate	where update A till convergence within each iteration (default: F, recommended: F)
beta_expect_flag	you can specify a fixed beta_expect by assigning a value, e.g., 1e-2 (default: NULL)

Value

X	same as input
K	same as input
M	same as input
beta_threshold	same as input
S_threshold	same as input
rho_S_diff	same as input
phi_S_prior	same as input
a_S_prior	same as input
b_S_prior	same as input
a_noise_prior	same as input
b_noise_prior	same as input
lambda_A_prior	same as input
eta_A_prior	same as input
loop	number of iterations before converge
S_expect	expectation of latent factors S, see the matching table
S_2_expect	see the matching table
mu_S_expect	see the matching table
epsilon_expect	see the matching table
sigma_log_S_expect	see the matching table
beta_expect	see the matching table
sigma_expect	see the matching table
mu_S_2_expect	see the matching table
mean_A_expect	expectation of loading A, see the matching table
var_A_expect	see the matching table
sigma_S	see the matching table
mu_S	see the matching table
lambda_S	see the matching table
pi_S	see the matching table
rho_S	see the matching table
phi_S	see the matching table
a_S	see the matching table
b_S	see the matching table
a_noise	see the matching table
b_noise	see the matching table
lambda_A	see the matching table
eta_A	see the matching table
mu_A	see the matching table
sigma_A	see the matching table
rho_S_0	initial values of rho_S
phi_S_0	initial values of phi_S
a_S_0	initial values of a_S
b_S_0	initial values of b_S
lambda_S_0	initial values of lambda_S
pi_S_0	initial values of pi_S
a_noise_0	initial values of a_noise
b_noise_0	initial values of b_noise
lambda_A_0	initial values of lambda_A
eta_A_0	initial values of eta_A
mu_S_change	relative change of mu_S, normF(mu_S-mu_S_old)/normF(mu_S_old)
beta_change	relative change of beta
ELBO_accu	ELBO value

Examples

NIFA.res <- NIFA(tscale(rnaseq.norm), K = 25, M = 4, max.iter = 500, S_threshold = 6e-5, init = "sd", 
                 ref = one_hot_encode(celltype), b_noise_prior = prod(dim(rnaseq.norm))*5)

wgmao/NIFA documentation built on March 31, 2022, 2:10 a.m.