Skip to contents

Performing Niche-DE

Performing_Niche_DE.Rmd

Running Niche-DE

Once you have set up your niche-DE object, you can run niche-DE using the function ‘niche_DE’. This function takes 11 arguments

Arguments
  • object: A niche-DE object
  • num_cores: The number of cores to use when performing parallelization
  • outfile: The file path to the txt on which workers will write status reports on
  • C: The minimum total expression of a gene across observations needed for the niche-DE model to run. The default value is 150.
  • M: Minimum number of spots containing the index cell type with the niche cell type in its effective niche for (index,niche) niche patterns to be investigated. The default value is 10
  • Gamma: Percentile a gene needs to be with respect to expression in the index cell type in order for the model to investigate niche patterns for that gene in the index cell. The default value is 0.8 (80th percentile)
  • print: Should the progress of niche-DE be tracked? If True, progress updates will be tracked in the outfile of the cluster. Default is True.
  • Int: Logical for if data is count data. If True, a negative binomial regression is performed. Otherwise linear regression with a gene specific variance is applied. Default is True.
  • Batch: Logical if an indicator should be included for each batch in the regression. Default is True.
  • self_EN: Logical if niche interactions between the same cell types should be considered. Default is False.
  • G: Number of gigabytes each core should hold. If the counts matrix is bigger than G gigabytes, it is split into multiple chunks such that the size of each chunk is less than G gigabytes.
#Perform Niche-DE
NDE_obj = niche_DE(NDE_obj,num_cores = 4,outfile = "",C = 150, M = 10, gamma = 0.8,print = T, Int = T, batch = T,self_EN = F,G = 1)

For some smaller datasets, it may be faster to not parallelize at all. You can use the “niche_DE_no_parallel” function in this case. It takes the same arguments sans num_cores,outfile, and G.

#Perform Niche-DE
NDE_obj = niche_DE_no_parallel(NDE_obj,C = 150, M = 10, gamma = 0.8,print = T, Int = T, batch = T,self_EN = F)
What does the output look like?

After running niche-DE, the ‘niche-DE’ slot in your niche-DE object will be populated. It will be a list with length equal to the length of sigma. Each item of that list will have an entry corresponding to each gene. Each entry is a list with 4 items. For a given gene k.

  • T-stat: A matrix of dimension #cell types by #cell types. Index (i,j) represents the T_statistic corresponding to the hypothesis test of testing whether gene k is an (index cell type i, niche cell type j) niche gene.
  • Beta: A matrix of dimension #cell types by #cell types. Index (i,j) represents the beta coefficient corresponding to the niche effect of niche cell type j on index cell type i for gene
  • nulls: Which coefficients are null (no niche-DE effect computed) for gene k.
  • var-cov: A matrix of dimension (#non nulls) by (#non nulls) . The matrix gives the variance covariance matrix of the beta coefficients of the niche-DE model for gene k.To save memory, this matrix is upper triangular.
  • log-lik: The log-likelihood of the niche-DE model for gene k.

Note that each item in the niche-DE list is named based on an element of sigma and the T-stat,beta,var-cov,log-lik items for that list are based on an effective niche calculated using a kernel bandwidth equal to that element of sigma. Additionally, the following two slots in your niche-DE object will be populated

  • Niche-DE-pval-pos: Pvalues for testing if a gene is an (index,niche)+ niche gene. This is a list with length equal to the length of sigma. Each sublist contains 3 items.
    • gene-level: A list of gene level pvalues. It is a vector with length equal to the number of genes.
    • cell-type-level: A matrix of dimension #genes by #cell types which gives cell type level pvalues.Index (i,j) gives a pvalue corresponding to whether gene i is a niche gene for index cell type j.
    • interaction-level: An array of dimension #cell types by #cell types by #genes which gives interaction level pvalues. Index (i,j,k) gives a pvalue corresponding to whether gene k is an (index cell type i, niche cell type j)+ niche gene.
  • Niche-DE-pval-neg: Pvalues for testing if a gene is an (index,niche)- niche gene. This is a list with length equal to the length of sigma. Each sublist contains 3 items.
    • gene-level: A list of gene level pvalues. It is a vector with length equal to the number of genes.
    • cell-type-level: A matrix of dimension #genes by #cell types which gives cell type level pvalues.Index (i,j) gives a pvalue corresponding to whether gene i is a niche gene for index cell type j.
    • interaction-level: An array of dimension #cell types by #cell types by #genes which gives interaction level pvalues. Index (i,j,k) gives a pvalue corresponding to whether gene k is an (index cell type i, niche cell type j)- niche gene.