compute.atac.network: ATAC-seq Gene regulatory network

In iaconogi/bigSCale2: A framework for the analysis of single cell data (RNA-seq and ATAC-seq) including clustering,phenotyping,pseudotime and inferring gene regulatory networks.

Description

Infers the gene regulatory network from single cell ATAC-seq data

Usage

compute.atac.network(expr.data, feature.file, quantile.p = 0.998)

Arguments

expr.data	matrix of expression counts. Works also with sparse matrices of the Matrix package.
quantile.p	only the first 1 - quantile.p correlations are used to create the edges of the network. If the networ is too sparse(dense) decrease(increase) quantile.p
gene.names	character of gene names, now it supports Gene Symbols or Ensembl, Mouse and Human.
clustering	type of clustering and correlations computed to infer the network. recursive Best quality at the expenses of computational time. If the dataset is larger than 10-15K cells and is highly heterogeneous this can lead to very long computational times (24-48 hours depending of the hardware). direct Best trade-off between quality and computational time. If you want to get a quick output not much dissimilar from the top quality of recursive one use this option. Can handle quickly also large datasets (>15-20K cells in 30m-2hours depending on hardware) normal To be used if the correlations (the output value cutoff.p) detected with either direct or recursive are too low. At the moment, bigSCale displays a warning if the correlation curoff is lower than 0.8 and suggests to eithe use normal clustering or increase the input parameter quantile.p
speed.preset	Used only if clustering='recursive' . It regulates the speed vs. accuracy of the Zscores calculations. To have a better network quality it is reccomended to use the default slow. slow Highly reccomended, the best network quality but the slowest computational time. normal A balance between network quality and computational time. fast Fastest computational time, worste network quality.
previous.output	previous output of compute.network() can be passed as input to evaluate networks with a different quantile.p without re-running the code. Check the online tutorial at https://github.com/iaconogi/bigSCale2.

Value

A list with the following items:

• centrality Main output: a Data-frame with the network centrality (Degree,Betweenness,Closeness,PAGErank) for each gene(node) of the network

• graph The regulatory network in iGraph object

• correlations All pairwise correlations between genes. The correlation is an average between Pearson and Spearman. Note that it is stored in single precision format (to save memory space) using the package float32.To make any operation or plot on the correlations first transform it to the standard double precisione by running correlations=dbl(correlations)

• cutoff.p The adptive cutoff used to select significant correlations

• tot.scores The Z-scores over which the correlations are computed. The visually check the correlation between to genes i and j run plot(tot.scores[,i],tot.scores[,j])

• clusters The clusters in which the cells have been partitioned

• model Bigscale numerical model of the noise

Examples

out=compute.network(expr.data,gene.names)

iaconogi/bigSCale2 documentation built on July 18, 2020, 8:21 p.m.