Extract Motif PFMs for MotifMatrix features

[jindalk]

Hi,
Is it possible to extract Position Frequency Matrices for the 884 motifs used for the chromVAR analysis? I am trying to run Tomtom for some of the motif targets that I have identified in my dataset: https://meme-suite.org/meme/tools/tomtom

Thanks,
Kunal

[Zepeng-Mu]

Maybe this #1226?

[jindalk]

Ah thanks! One more question, since I need to convert these PWMs to PFMs for Tomtom, would you know what pseudocount was used to obtain these PWMs? I dug around a bit and found this in the original chromVAR paper - These position frequency matrices were then converted into position weight matrices (PWMS) by taking the log of the frequency after adding a 0.008 pseudocount and dividing by 0.25.

I wonder if the same params were used in ArchR motif models as well?

[Zepeng-Mu]

I think (I could be wrong) ArchR is using 0.008 as well, and this is also the default used in TFBSTools package.

[rcorces]

I also believe this is correct.

[jindalk]

Got it, thank you both!

[jindalk]

Hi,
I tried the method that was suggested to extract PFMs, but I am not sure if I'm still getting the correct matrix. This is the code I used:

all_motifs <- b4.atac@peakAnnotation[['Motif']]$motifs
foxa1_motif <- TFBSTools::as.matrix(all_motifs[['Foxa1_306']])
foxa1_pfm <- 0.25*(2**foxa1_motif) - 0.008

In the final matrix, when I compute colSums, it is not equal to ~1 for each of the positions. It is actually as low as 0.7 in some cases.
colSums(foxa1_pfm)
0.9484407 0.8164749 0.8840974 0.8772556 0.7316546 0.7721702 0.7592677 0.7276554 0.6784628 0.7938169 [11] 0.6837447 0.9599072 0.9553515 0.9654302 0.9576198

Shouldn't colSums of PFM by definition be equal to 1, unless I am understanding this incorrectly?

Thanks

[rcorces]

maybe this helps #476 (comment)

[jindalk]

The issue linked below that comment (#509) solved it for me. I was using log 2 but it had to be natural log instead. Thank you!