Filtering content in heatmap to highest values

Question

[cthangav]

I have a dataframe called "ListerTFHeat" that is 20,000 by 20,000. First few rows/columns below. I want to pull out the top twelve rows and columns to create a simple heatmap, instead of plotting the whole thing.

ListerTFHeat[1:20,1:20]
       TFName 0610009B22Rik 0610009O20Rik 0610010F05Rik 0610010K14Rik 0610030E20Rik 0610037L13Rik 0610040J01Rik 1110004E09Rik
  1:      Afp      0.000000             0        0.0000             0      0.000000       0.00000      0.000000       0.00000
  2:      Ahr      0.000000             0        0.0000             0      4.293800       0.00000      0.000000       0.00000
  3:     Aire      0.000000             0        0.0000             0      0.000000       0.00000      0.000000       0.00000
  4:     Alx1     10.212000             0      128.1400             0      3.775000      16.15100      8.031200      70.98600
  5:     Alx3      0.089674             0        1.1713             0      0.033633       0.14696      0.075401       0.61377
  6:     Alx4      8.916400             0      119.6300             0      3.451800      14.90500      7.940900      61.04400
  7:       Ar    159.390000             0      602.0300             0     78.846000      22.38300      4.474100     460.23000
  8: Arhgef12      0.000000             0        0.0000             0      0.000000       0.00000      0.000000       0.00000
  9:   Arid3a      0.000000             0        0.0000             0      0.000000       0.00000      0.000000       0.00000
 10:   Arid3b      0.000000             0        0.0000             0      0.000000       0.00000      0.000000       0.00000
 11:   Arid5a      0.000000             0        0.0000             0      0.000000       0.00000      0.000000       0.00000
 12:   Arid5b      0.000000             0        0.0000             0      0.000000       0.00000      0.000000       0.00000
 13:     Arnt     23.332000             0        4.2158             0     18.824000       0.00000      0.000000       0.00000
 14:    Arnt2     68.152000             0       13.3690             0     34.317000       0.00000      0.000000       0.00000
 15:    Arntl      0.000000             0        0.0000             0      0.000000       0.00000      0.000000       0.00000
 16:   Arntl2      0.000000             0        0.0000             0      0.000000       0.00000      0.000000       0.00000
 17:      Arx      1.815100             0       28.8670             0      0.804270       2.97110      1.669500      13.50500
 18:    Ascl1     12.171000             0       74.4150             0      5.630500      14.19200      0.000000       0.00000
 19:    Ascl2     73.566000             0      438.6500             0     38.658000      91.22500      0.000000       0.00000
 20:     Atf1      0.000000             0      185.5600             0     13.039000       0.00000      0.000000      35.42100

If there is some sort of function that can extract a subset of the data that would result in a heatmap with the highest values, that would be ideal.

If not, (I think) the next best would be taking the top 12 rowsums and all columns. Then of those columns, selecting the top 12 colsums. Then creating the heatmap.

What is a way to code this in R? I haven't been able to find a post for this because most of them have been about scaling/sizing a heatmap.

[Papyrus]

You mean something like this? depends on how you define "top" row or column

# Create example (it is a data.table)
ListerTFHeat <- data.table::data.table(
  TFName = sample(LETTERS, 20, F)
)
ListerTFHeat <- cbind(ListerTFHeat,replicate(10,rnorm(10, 10, 10)))

# Converte to matrix with rownames
mt <- as.matrix(ListerTFHeat[,-1])
rownames(mt) <- ListerTFHeat$TFName

# Order by rowSums, colsums
mt.ordered <- mt[order(rowSums(mt), decreasing = T),order(colSums(mt), decreasing = T)]
mt.ordered[1:3,1:3]