Filter columns by group and condition

Question

[Sissi]

Hi there,

I have a kind of easy task but still can't figure it out. I have a csv binary matrix, with genes as rows and samples as columns, like this:

Gene	sampleA	sampleB	sampleC	sampleD	sampleE	sampleF	sampleG
gene1	1	0	0	1	0	0	0
gene2	0	0	0	0	1	1	0
gene3	0	0	0	0	0	0	1
gene4	0	1	0	0	0	0	0
gene5	1	1	1	1	0	0	0
gene6	1	1	1	1	0	0	0
gene8	0	0	0	0	0	0	1
gene9	0	0	0	0	0	0	0
gene10	1	0	0	1	0	0	0
gene11	0	0	0	0	1	1	1
gene12	0	0	0	0	1	1	1
gene13	0	0	0	0	0	0	0
gene14	0	0	0	0	0	1	0
gene16	1	0	0	0	0	0	0
gene17	1	0	0	1	0	0	0
gene18	1	0	0	1	0	0	0
gene19	1	0	0	1	0	0	0
gene20	1	0	0	1	0	0	0

The samples belong to specific clusters, like:

cluster1 = c(sampleA, sampleB, sampleC, sampleD)
cluster2 = c(sampleE, sampleF, sampleG)

I would like to subset/filter the columns according to the gene presence in only one cluster, like this:

Gene	sampleA	sampleB	sampleC	sampleD	sampleE	sampleF	sampleG
gene5	1	1	1	1	0	0	0
gene6	1	1	1	1	0	0	0
gene11	0	0	0	0	1	1	1
gene12	0	0	0	0	1	1	1

In a way to see which gene is present only in one of the two clusters.

Is there any easy way with R or bash?

Thanks

[Ram]

Add a column for cluster and use group by and sum across all but the gene and cluster columns. That'd be the general approach I'd take, you should be able to figure out specific dplyr functions with that outline in mind.

[Sissi]

Hi Ram, Thanks. But then the cluster column would refers to the gene column and not to the sample.

[rpolicastro]

You need to convert your data to long format first using e.g. pivot_longer from tidyr.

Answer 1

[Matthias Zepper]

If you just have two clusters, you can also run a summarization function first:

# simulate data
binary_matrix <- matrix(rbinom(1e4,1,0.6),ncol=10)
colnames(binary_matrix) <- paste0("sample",LETTERS[1:10])
rownames(binary_matrix) <- paste0("gene",c(1:1000))

expressed_clusterA <- apply(binary_matrix[,1:5],1,all)
expressed_clusterB <- apply(binary_matrix[,6:10],1,any)

#get those genes that are expressed in all A samples, but not in B
binary_matrix[expressed_clusterA & !expressed_clusterB,]

You can apply arbitrary and custom functions like this, not only any(), all() etc.

apply(binary_matrix[,1:5],1,function(x){all(abs(diff(x))==1)})

for example finds genes with alternating expression within clusterA (that is either 0,1,0,1,0 or 1,0,1,0,1)

Like so, you can flexibly configure the search patterns you are looking for. Use sum() for conditions like "expressed in at least 4 out of 5" etc.

[Sissi]

That perfectly worked. Thank you!