runLDA: Latent Dirichlet Allocation

In r3fang/SnapATAC: Single Nucleus Analysis Package for ATAC-Seq

Latent Dirichlet Allocation

Description

This function takes a Snap obj as input with bmat/pmat slot and run Latent Dirichlet Allocation (LDA).

Usage

runLDA(obj, input.mat = c("bmat", "pmat"), topic = c(10, 20, 30),
  method = c("Z-score", "Probability"), num.cores = 1, min.cell = 10,
  seed.use = 10, iterations = 500, burnin = 250, alpha = 50,
  alphaByTopic = TRUE, beta = 0.1)

Arguments

obj	A snap obj
input.mat	Input matrix to be used for LSA c("bmat", "pmat").
topic	An integer number of topics to return c(10, 20, 30).
method	Method used to normalize the cell-by-topic score c("Z-score", "Probability").
num.cores	Number of cores used for computing [1].
min.cell	Min cell coverage. Features with coverage less than min.cell will be filtered [10].
seed.use	A numeric class that indicates random seeding number [10].
iterations	The number of sweeps of Gibbs sampling over the entire corpus to make.
burnin	A scalar integer indicating the number of Gibbs sweeps to consider as burn-in (i.e., throw away) for 'lda.collapsed.gibbs.sampler' and 'mmsb.collapsed.gibbs.sampler'. If this parameter is non-NULL, it will also have the side-effect of enabling the document_expects field of the return value (see below for details). Note that burnin iterations do NOT count towards num.iterations.
alpha	The scalar value of the Dirichlet hyperparameter for topic proportions.
alphaByTopic	scale alpha by topic number
beta	The scalar value of the Dirichlet hyperparamater for topic multinomials.

Details

LDA is a generative statistical model that allows sets of observations to be explained by unobserved groups that explain why some parts of the data are similar. was first applied to analyze single cell ATAC-seq by Cis-Topic (González-Blas, Nature Methods, 2019). LDA iteratively optimize two probability distributions: (1) the probability of a region belonging to a topic (region–topic distribution) and (2) the contribution of a topic within a cell (topic–cell distribution).

Multiple LDA models will be trained and the optimal one will be selected according to the likelohood.

Examples

data(demo.sp);
demo.sp = makeBinary(demo.sp);
demo.sp = runLDA(
obj=demo.sp, 
input.mat="bmat", 
topic=c(10, 20, 30), 
method="Z-score", 
min.cell=0,
num.cores=3
);

r3fang/SnapATAC documentation built on March 29, 2022, 4:33 p.m.