hicConvertFormat : Missing negative values in ginteractions (.tsv) output

[u-n-i-v-e-r-z]

Hi everyone,

I hope you're all fine.
I noticed that hicConvertFormat won't return negative values that are still present in a given .h5 matrix (norm KR).
This matrix represent differential interaction measures between two given conditions.

I'm using hicexplorer_3.4.3
Here is the output when I import the .h5 matrix :

>>> ma = HiCMatrix.hiCMatrix("myMatrix_25kb.norm.KR.h5")   
>>> ma.getMatrix()

matrix([[ 1.8099686 , -1.33049537, -3.69973388, ...,  0.        ,
          0.        , -1.08915401],
        [-1.33049537,  2.01023492, -3.59139214, ..., -1.1260478 ,
          0.16709778,  0.20518745],
        [-3.69973388, -3.59139214,  1.7171135 , ..., -0.7801746 ,
          0.        , -0.19838392],
        ...,
        [ 0.        , -1.1260478 , -0.7801746 , ...,  1.69022627,
          1.68831546, -0.23708201],
        [ 0.        ,  0.16709778,  0.        , ...,  1.68831546,
          0.96983819, -0.15338982],
        [-1.08915401,  0.20518745, -0.19838392, ..., -0.23708201,
         -0.15338982,  3.50339214]])

Negative values are present and it is confirmed when plotting the matrix.

Then I use hicConvertFormat like this :

hicConvertFormat --matrices myMatrix_25kb.norm.KR.h5 --outFileName myMatrix_25kb.norm.KR.gi --inputFormat h5 --outputFormat ginteractions

And this is the output of the .tsv file I obtain (under R in cm object):

> cm <- data.table::fread("myMatrix_25kb.norm.KR.gi.tsv")
> cm[order(cm$V4,cm$V5),]

   V1       V2       V3 V4       V5       V6         V7
      1:  I        0    25000  I        0    25000 1.80996860
      2:  I    25000    50000  I    25000    50000 2.01023492
      3:  I    50000    75000  I    50000    75000 1.71711350
      4:  I    50000    75000  I    75000   100000 0.04064520
      5:  I    75000   100000  I    75000   100000 1.71038424

What is weird is when asking for negative values, the only one I can obtain is located at the very end of the chromosome I:

cm[cm$V7 < 0,]

V1       V2       V3 V4       V5       V6         V7
1:  I 15050000 15072434  I 15050000 15072434 -0.6984005

I would rather use .h5 file but I have several problems with blosc filter compatibility and I can't load matrices under R currently.
Any workaround would be greatly appreciated !

BR

Alex

[joachimwolff]

The posted values are matching in the matrix and in your ginteractions file. At least the first three, these are exactly the diagonal values (0, 0), (1,1) and (2,2).

• What is the content of the ginteractions file if you do not open it with R, but sth like less?
• What are the values at I 0 25000 I 25000 50000, I 0 25000 I 50000 75000, I 0 25000 I 75000 100000?

We just write out what the content of the matrix is: https://github.com/deeptools/HiCMatrix/blob/master/hicmatrix/lib/ginteractions.py#L25

Maybe the problem is located in R?

[u-n-i-v-e-r-z]

@joachimwolff,

- What is the content of the ginteractions file if you do not open it with R, but sth like less?

Those are the first lines from call to "less" :

I       0       25000   I       0       25000   1.8099686004600926
I       0       25000   I       3675000 3700000 0.2824356663718901
I       0       25000   I       3975000 4000000 0.46480353773800975
I       0       25000   I       4000000 4025000 0.136099560651966
I       0       25000   I       4575000 4600000 0.35421538999192403
I       0       25000   I       5100000 5125000 0.606264897271899

- What are the values at I 0 25000 I 25000 50000, I 0 25000 I 50000 75000, I 0 25000 I 75000 100000?

> cm[V1=="I" & V2==0 & V3==25000 & V4=="I" & V5==25000 & V6==50000,]
Empty data.table (0 rows and 7 cols): V1,V2,V3,V4,V5,V6...

> cm[V1=="I" & V2==0 & V3==25000 & V4=="I" & V5==50000 & V6==75000,]
Empty data.table (0 rows and 7 cols): V1,V2,V3,V4,V5,V6...

> cm[V1=="I" & V2==0 & V3==25000 & V4=="I" & V5==75000 & V6==100000,]
Empty data.table (0 rows and 7 cols): V1,V2,V3,V4,V5,V6...

> # Test on random ranges diagonal
> cm[V1=="I" & V2==0 & V3==25000 & V4=="I" & V5==0 & V6==25000,]
   V1 V2    V3 V4 V5    V6       V7
1:  I  0 25000  I  0 25000 1.809969

> # Test on random ranges out diag
> cm[V1=="I" & V2==50000 & V3==75000 & V4=="I" & V5==75000 & V6==100000,]
   V1    V2    V3 V4    V5     V6        V7
1:  I 50000 75000  I 75000 100000 0.0406452

To me nothing is wrong here, negative values are simply discarded.
To be sure I tested if I had the same number of lines in raw .tsv and after loading it into R :

Bash

ẁc -l myMatrix_25kb.norm.KR.gi.tsv
3091491

R

> nrow(cm)
[1] 3091491