How to reduce a massive matrix conditioned on a specific variable that does not contain NA-CodePudding

I have an enormous "omics" dataset containing three different experiments: df$method == Mut, Spy, VAR

  method    a    b    c    d
1    Mut 12.3   NA   NA 17.5
2    Spy 13.5   NA   NA   NA
3    VAR 13.2 19.6 11.1   NA
4    Mut   NA   NA   NA   NA
5    Spy   NA   NA   NA 19.9
6    VAR   NA 20.1 18.6   NA

Using dplyr, how can I reduce the matrix so it only contains rows where df$method == VAR has values (at least one value)? I.e., where all values in a, b, c, d ... is NA for df$method == Mut, Spy.

Shown on a Venn Diagramm, values that fits in the white area, are of interest.

So, the expected output from df would be:

> df
  method    b    c
1    VAR 19.6 11.1
2    VAR 20.1 18.6

Data

df <- structure(list(method = c("Mut", "Spy", "VAR", "Mut", "Spy",
                                "VAR"), a = c(12.3, 13.5, 13.2, NA, NA, NA), b = c(NA, NA, 19.6,
                                                                                   NA, NA, 20.1), c = c(NA, NA, 11.1, NA, NA, 18.6), d = c(17.5,
                                                                                                                                           NA, NA, NA, 19.9, NA)), class = "data.frame", row.names = c(NA,
                                                                                                                                                                                                       -6L))

CodePudding user response：

dplyr option to first filter the method and then select the columns with no NA's like this:

df <- structure(list(method = c("Mut", "Spy", "VAR", "Mut", "Spy",
                                "VAR"), a = c(12.3, 13.5, 13.2, NA, NA, NA), b = c(NA, NA, 19.6,
                                                                                   NA, NA, 20.1), c = c(NA, NA, 11.1, NA, NA, 18.6), d = c(17.5,
                                                                                                                                           NA, NA, NA, 19.9, NA)), class = "data.frame", row.names = c(NA,
                                                                                                                                                                                                       -6L))
library(dplyr)
library(dplyr)   
df %>%
  filter(method == "VAR") %>%
  select_if(~!any(is.na(.)))
#>   method    b    c
#> 1    VAR 19.6 11.1
#> 2    VAR 20.1 18.6

^{Created on 2022-07-06 by the reprex package (v2.0.1)}

CodePudding user response：

Here is a base R way. Use logical indices to get the rows where method == "VAR" and the columns where the other rows, the rows with method is equal to "Spy" or "Mut" are all NA.

df <- structure(list(
  method = c("Mut", "Spy", "VAR", "Mut", "Spy","VAR"), 
  a = c(12.3, 13.5, 13.2, NA, NA, NA), 
  b = c(NA, NA, 19.6,NA, NA, 20.1), 
  c = c(NA, NA, 11.1, NA, NA, 18.6), 
  d = c(17.5,NA, NA, NA, 19.9, NA)), 
  class = "data.frame", row.names = c(NA,-6L))

i_row <- df$method == "VAR"
i_col <- colSums(is.na(df[!i_row, -1])) == nrow(df[!i_row,])
df[i_row, c(TRUE, i_col)]
#>   method    b    c
#> 3    VAR 19.6 11.1
#> 6    VAR 20.1 18.6

^{Created on 2022-07-06 by the reprex package (v2.0.1)}