I have 6 matrices and I have to perform a t-test over cell values combined from the 6 matrices and store the p-value in a new matrix

Question

Input: There are 6 input matrices of same dimensions 3 input matrices from normal tissues:

              GeneA             GeneB          
GeneA          31                  4           
GeneB           5                  8 

              GeneA             GeneB          
GeneA           5                 14           
GeneB           5                  8 


              GeneA             GeneB          
GeneA          30                 14           
GeneB           45                 7

3 input matrices from cancer tissues:

              GeneA             GeneB          
GeneA          11                  4           
GeneB           5                  18 

              GeneA             GeneB          
GeneA           7                 14           
GeneB           15                 4 


              GeneA             GeneB          
GeneA          30                 14           
GeneB           45                 7

output:

                  GeneA                            GeneB          
GeneA        t-test({31,5,30},{11,7,13})    t-test({4,14,14},{4,14,14})        
GeneB        t-test({5,5,45},{5,15,45})     t-test({8,8,7},{18,4,7})

Output matrix will have the p-values from the test

What are you testing, what do you want to highlight? What do contigency tables represent? — Yacine Hajji, Jun 01 '22 at 08:02
I m testing gene expression across normal and cancer tissue types — Nikita Srivastav, Jun 01 '22 at 08:10
Please post data in `dput` format. Do the matrices have names following a pattern? Even if two groups of patterns, for normal and cancer data? — Rui Barradas, Jun 01 '22 at 08:29
No there is no such pattern, but I can change their names to create a pattern over the two groups — Nikita Srivastav, Jun 01 '22 at 08:41

score 1 · Answer 1 · answered Jun 01 '22 at 09:05

The code that follows conducts the t-tests on data in a 3-dim array form. This makes it easier to loop through the data sets, extract the required vectors and run the tests.

From tabular data to arrays

normal <- mget(ls(pattern = "^normal"))
cancer <- mget(ls(pattern = "^cancer"))

anorm <- array(dim = c(dim(normal[[1]]), length(normal)))
acanc <- array(dim = c(dim(cancer[[1]]), length(cancer)))
for(i in seq_along(normal)) {
  anorm[, , i] <- unlist(normal[[i]])
  acanc[, , i] <- unlist(cancer[[i]])
}

The t-tests

Create a results list first, then run the tests, then extract the p-values into a data.frame.

t_test_list <- vector("list", length = prod(dim(anorm)[1:2]))
for(j in seq(dim(anorm)[2])) {
  for(i in seq(dim(anorm)[1])) {
    x <- anorm[i, j, ]
    y <- acanc[i, j, ]
    k <- i + dim(anorm)[1]*(j - 1)
    t_test_list[[k]] <- t.test(x, y)
  }
}

t_test_pval <- normal[[1]]
t_test_pval[] <- sapply(t_test_list, `[[`, 'p.value')

t_test_pval
#>           GeneA     GeneB
#> GeneA 0.6176386 1.0000000
#> GeneB 0.8618120 0.6850214

^{Created on 2022-06-01 by the reprex package (v2.0.1)}

Data

x<-'              GeneA             GeneB          
GeneA          31                  4           
GeneB           5                  8'
normal1 <- read.table(textConnection(x), header = TRUE)

x<-'              GeneA             GeneB          
GeneA           5                 14           
GeneB           5                  8 '
normal2 <- read.table(textConnection(x), header = TRUE)


x<-'              GeneA             GeneB          
GeneA          30                 14           
GeneB           45                 7 '
normal3 <- read.table(textConnection(x), header = TRUE)

x<-'              GeneA             GeneB          
GeneA          11                  4           
GeneB           5                  18 '
cancer1 <- read.table(textConnection(x), header = TRUE)

x<-'              GeneA             GeneB          
GeneA           7                 14           
GeneB           15                 4 '
cancer2 <- read.table(textConnection(x), header = TRUE)


x<-'              GeneA             GeneB          
GeneA          30                 14           
GeneB           45                 7 '
cancer3 <- read.table(textConnection(x), header = TRUE)

^{Created on 2022-06-01 by the reprex package (v2.0.1)}

score 1 · Answer 2 · answered Jun 01 '22 at 09:08

Perhaps the neatest way to this is to bind both sets of matrices into 2 x 2 x 3 arrays, then use Map to get the t tests at each of the four combinations.

Let's say your matrices are called norm1, norm2, norm3 for the normal tissue and ca1, ca2 and ca3 for the cancer tissue. Then we can do:

library(abind)

norm <- abind::abind(norm1, norm2, norm3, along = 3)
canc <- abind::abind(ca1, ca2, ca3, along = 3)

pvals <- Map(function(i, j) {
  t.test(norm[i, j, ], canc[i, j, ])$p.val
  }, i = c(1:2, 1:2), j = c(1, 1, 2, 2))

matrix(unlist(pvals), 2, dimnames = dimnames(norm1))
#>           GeneA     GeneB
#> GeneA 0.6176386 1.0000000
#> GeneB 0.8618120 0.6850214

^{Created on 2022-06-01 by the reprex package (v2.0.1)}

Reproducible data

norm1 <- structure(c(31L, 5L, 4L, 8L), dim = c(2L, 2L), dimnames = list(
    c("GeneA", "GeneB"), c("GeneA", "GeneB")))

norm2 <- structure(c(5L, 5L, 14L, 8L), dim = c(2L, 2L), dimnames = list(
    c("GeneA", "GeneB"), c("GeneA", "GeneB")))

norm3 <- structure(c(30L, 45L, 14L, 7L), dim = c(2L, 2L), dimnames = list(
    c("GeneA", "GeneB"), c("GeneA", "GeneB")))

ca1 <- structure(c(11L, 5L, 4L, 18L), dim = c(2L, 2L), dimnames = list(
    c("GeneA", "GeneB"), c("GeneA", "GeneB")))

ca2 <- structure(c(7L, 15L, 14L, 4L), dim = c(2L, 2L), dimnames = list(
    c("GeneA", "GeneB"), c("GeneA", "GeneB")))

ca3 <- structure(c(30L, 45L, 14L, 7L), dim = c(2L, 2L), dimnames = list(
    c("GeneA", "GeneB"), c("GeneA", "GeneB")))

Instead of abind, as it only takes 3 matrices, can i use something else as the number of matrices may increase to 5 — Nikita Srivastav, Jun 01 '22 at 09:21
Also, my matrices are sized 1000*1000 each, so for the map function pvals <- Map(function(i, j) { t.test(norm[i, j, ], canc[i, j, ])$p.val }, i = c(1:1000, 1:1000), j = c(1, 1, 2, 2)) — Nikita Srivastav, Jun 01 '22 at 09:24
For matrices 1000 * 1000 you would have `i = rep(1:1000, 1000)` and `j = rep(1:1000, each = 1000)` — Allan Cameron, Jun 01 '22 at 09:38

I have 6 matrices and I have to perform a t-test over cell values combined from the 6 matrices and store the p-value in a new matrix

2 Answers2

From tabular data to arrays

The t-tests

Data