InferCNV

If you wanted to run InferCNV using Betsy:

GTF=/data/genomidata/genomes/Broad.hg19.RefSeq.NM_only.170811.gtf
betsy_run.py --num_cores 10 \
--input SignalFile --input_file counts.txt \
--dattr SignalFile.preprocess=counts \
--input GTFGeneModel --input_file ${GTF} \
--output InferCNVAnalysis --output_file output_folder \
--mattr reference_dataset=ref.counts.1.txt \
--mattr reference_dataset2=ref.counts.2.txt \
--mattr reference_name=ref.name.1 \
--mattr reference_name2=ref.name.1 \
--run

alternatives
If you do not want to generate plots--substitute InferCNVAnalysis with InferCNVOutput.

If you wanted to run within R

input files

fibroblasts counts is used as a reference here.

• filtered counts (the "filtered_counts" in the demo script):

  the counts of the high quality cells in the format of tab-delimited plain text file, looking like so:

Gene.Symbol	Cell1	Cell2	...
A	1	2	...
B	0	5	...
C	2	6	...
...	...	...	...

• reference counts: a plain text file looking like so:

Gene.Symbol	Ref1	Ref2	...
C	1	0	...
A	0	0	...
B	0	1	...
...	...	...	...

• cell annotations (the "singler_annot" in the demo script):

  a plain text file with column headers:

Cell annot
Cell1 Epithelial
Cell2 Endotheial
... ...

• gene position file (the "gene_order_file" in the "CreateInfercnvObject" function):

  The format is tab-delimited and has no column header, simply providing the gene name, chromosome, and gene span:

A    Chr2	14363	27239
B    ChrX 761586 762902
C    Chr5 1152288 1167411
...    ...

scripts

# this script is reproducible in the R docker image on unicron

# load librarys
library(infercnv)
library(dplyr)
library(data.table)

set.seed(41)
# import raw data
## counts
### filtered counts
filtered_counts <- 
	fread('xxx/xxxfiltered_counts.txt', stringsAsFactors = F, check.names =F) %>% 
		as.data.frame
### fibroblast counts (reference)
fibro_counts <- 
	fread('xxx/80_scrna_cnv_normalization/Fibroblasts.counts.txt', stringsAsFactors = F,
        check.names =F) %>% 
		as.data.frame
## singler annot
singler_annot <- 
	read.delim('xxx/singler_filtered_annot.txt', stringsAsFactors = F, check.names = F)

# data cleaning
## change geneid colum name to genesymbol
colnames(fibro_counts)[1] = "Gene.Symbol"
## generate and write the annotation file
fibro_annot <- data.frame(Cell = colnames(fibro_counts)[-1], annot = "fibroblast")
write.table(rbind(fibro_annot, singler_annot), 
            "filtered_fibro-ref.txt", sep = "\t", quote = F, 
            col.names = F, row.names = F)
## generate counts matrix
filtered_fibro_counts_matrix <- 
inner_join(filtered_counts, fibro_counts, by = "Gene.Symbol") %>% 
	`rownames<-`(c(.$Gene.Symbol)) %>% 
	.[, !colnames(.) %in% c("Gene.Symbol")] %>% 
	as.matrix
 
# do infercnv
# creat infercvnobj
infercnvobj <- CreateInfercnvObject(raw_counts_matrix=filtered_fibro_counts_matrix, 
                                    annotations_file="filtered_fibro-ref.txt", delim="\t", 
                                    gene_order_file="xxx/gencode_v19_gene_pos.txt",
                                    ref_group_names=c("fibroblast"),  #an alternative: NULL
                                    chr_exclude = NULL #an alternative: c("ChrM", "ChrX")
                                   )

# perform infercnv operations to reveal cnv signal
infercnvobj = infercnv::run(infercnvobj, 
cutoff=0.1,
out_dir="./out-fibro-ref-k3",
cluster_by_groups=F, 
k_obs_groups = 3,
denoise=T, HMM=T,  
num_threads = 23, 
# if you want to get plots as one of the outputs, turn these two into F
## plotting steps takes pretty long time to finish, e.g., 200 cells dataset: "no plots" takes ~100sec, "with plots" takes >220sec.  
no_plot = T, no_prelim_plot = T
)

extract hclusts

cutree(x@tumor_subclusters$hc$all_observations, k = 4)	

#cr: Xuan 😊

References and documentation

Reference: Anoop P. Patel, Itay Tirosh, et al. Science. 2014 Jun 20: 1396-1401

Wiki: Broad Institute InferCNV home page (https://github.com/broadinstitute/inferCNV/wiki)

Package Manual