Non-negative Independent Factor Analysis for single cell RNA-seq

MASS
mclust
Rcpp
RcppArmadillo

library(devtools)
install_github("wgmao/NIFA")

The main function is NIFA() and there is a short vignette based on a test dataset (simulated scRNA-seq) called SimKumar4easy which is publicly available via the bioconductor package DuoClustering2018.

There are five parameters that are more sensitive than others: K, S_threshold/max.iter and b_noise_prior/beta_expect_flag.

• K number of latent factors.
• S_threshold and max.iter control the number of iterations.
• b_noise_prior Based on experience, the recommendation is to set it as prod(dim(X))*5). If the result doesn't look good, you can manually set up a fixed value for the noise parameter using beta_expect_flag This is highly data-dependent.