I have a dataset which contains count data where the structure looks like:

What I'm trying to get to with tidyr/dplyr is the per-sample correlation between each of the observed counts and the expected counts (i.e. I'm unfussed by the correlation between each of the observed columns).

I can do this by looping, but was wondering whether there is a 'clearer' approach to take using tidyverse functions?

Any help much appreciated!!

CodePudding user response：

How about this:

library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
library(tidyr)
d <- tibble::tribble(~Sample,   ~ID,    ~Expected,  ~Observed_A,    ~Observed_B,
"A",    "id1",  10, 8,  10,
"A",    "id2",  6,  8,  4,
"B",    "id1",  15, 12, 18,
"B",    "id2",  1,  2,  4)

d %>% 
  group_by(Sample) %>%
  summarise(as.data.frame(cor(Expected, cbind(Observed_A, Observed_B)))) %>% 
  pivot_longer(-Sample, names_to = "Dataset", values_to="Correlation")
#> Warning in cor(Expected, cbind(Observed_A, Observed_B)): the standard deviation
#> is zero
#> # A tibble: 4 × 3
#>   Sample Dataset    Correlation
#>   <chr>  <chr>            <dbl>
#> 1 A      Observed_A          NA
#> 2 A      Observed_B           1
#> 3 B      Observed_A           1
#> 4 B      Observed_B           1

^{Created on 2022-03-04 by the reprex package (v2.0.1)}

CodePudding user response：

df %>% 
  group_by(Sample) %>%
  summarize(across(Observed_A:Observed_B, ~cor(.x, Expected))) %>%
  pivot_longer(!Sample, values_to = "Correlation", names_to = "Dataset")