myNMF: Function to run NMF with different options

In naikai/sake: Single-cell RNA-Seq Analysis and Klustering Evaluation

Function to run NMF with different options

Description

This function allows you to perform matrix factorization using Non-negative matrix factorization (NMF) method User need to provide filtered/ordered gene expression data and client info (expinfo) files Workflow ### 1 estimate how many clusters (k) ### 1.1 estim.r <- nmf(data, 2:10, nrun=100) 1.1 random shuffle the row (genes) for each column (sample) - estim.random.r 1.2 generate plot(estim.r), consensusmap(estim.r), plot(estim.r, estim.random.r) 2 After looking through the graph and decide k # This can be automated? 2.1 res <- nmf(data, k, nrun=100?) 2.2 compare the original.data and fitted.data (heatmap.2) # ignore this. 2.3 summary (res), summary(res, target=data), summary(res, class=assigned.subtypes) 2.4 plot basismap(res) and coefmap(res) 2.5 extract metagene-specific features

Usage

myNMF(data, prefix = "NMF", cluster = 3, nrun = 100, norm = F,
  ncores = 8, algorithm = "brunet", mode = "real", seed = 123211)

Arguments

data	Input data sets
prefix	Prefix for file output. Default is 'NMF'
cluster	Estimated rank (clusters) in the data sets. Default is 3
algorithm	Which algorithms for NMF? (bruent, lee, nsNMF, KL, Frobenius, offset, ls-nmf, pe-nmf, siNMF). Default is brunet
mode	Which modules to run? (Estim or Real). Default is real
run	How many runs to perform? Default is 100

Examples

myNMF(data, mode="estim", cluster=3, nrun=20)

naikai/sake documentation built on Feb. 15, 2023, 11 p.m.