noMasked changes beta values

[NazaninMir]

Hi all,
I'm currently working on an EPICv2 array using sesame .. during the QC steps I realized that the beta-values change after applying the noMasked function on the SigDF. Anyone else experienced the same? or can someone explain this ?

[zwdzwd]

If you apply getBetas directly, like getBetas(noMasked(sdf)) and getBetas(sdf), there shouldn't be a difference. My guess is you are using openSesame which includes additional preprocessing some of which may use the masked probes.

[NazaninMir]

Thank you for your reply.
Actually I didn't used openSesame.
Thats my command:
sdfMask <- lapply(sdf, function (x) noMasked(x))

[zwdzwd]

Is sdf just one SigDF? under the hood noMasked is simply

sdf[!sdf$mask,,drop=FALSE]

so you just need

noMasked(sdf)

[NazaninMir]

No I have over 50 SigDF in sdf.
List of 56
$ 207222480102_R01C01:Classes ‘SigDF’ and 'data.frame': 937690 obs. of 7 variables:
....
I use lapply, cause I could not figure out wether other processes are running when using openSesame

[zwdzwd]

OK. then how did you check beta values and what else did you do before that?

[NazaninMir]

Here are the steps I applied. I did all QC step by step, to be able to reconstruct:
idatlist <- lapply(idats, function (x) readIDATpair(x, manifest = addr))
test.betas <- do.call(cbind, lapply(idatlist,function (x) getBetas(x, mask=TRUE)))

#inferInfiniumIChannel
idatlist_C <- lapply(idatlist, inferInfiniumIChannel)
test.betasC <- do.call(cbind, lapply(idatlist_C,function (x) getBetas(x, mask=TRUE)))

Run dye bias correction

idatlist_CD <- lapply(idatlist_C, dyeBiasNL)
test.betasCD <- do.call(cbind, lapply(idatlist_CD,function (x) getBetas(x, mask=TRUE)))

#Detection Pvalue
idatlist_CDPo <- lapply(idatlist_CD, function (x) pOOBAH(x, pval.threshold = 0.01))
test.betasCDPo <- do.call(cbind, lapply(idatlist_CDPo,function (x) getBetas(x, mask=TRUE)))

-> until here everything seems fine. Beta values are as they should be

#####################################################################

#-> after this step some beta values changes.

idatlist_CDPo_ExcMask <- lapply(idatlist_CDPo,function (x) noMasked(x))
test.betasCDPoMask <- do.call(cbind, lapply(idatlist_CDPo_ExcMask,function (x) getBetas(x, mask=TRUE)))

[zwdzwd]

thanks. looking at the code

sesame/R/sesame.R

Line 189 in 724d4c5

getBetas <- function(

The only change expected is the filtering of masked probes and hence shorter dimension but there should be no change to the value that survive masking. or am I missing something?

[NazaninMir]

There should not be a change in the values .. and that is also the case in all the steps before ....

test.betasCDPo[c("cgXXXX_BC11", "cgXXXY_TC21", "cgXXXZ_BC21"),][,1:3]
207222480102_R01C01 207222480102_R02C01 207222480102_R03C01
cg00054210_BC11 0.4165562 0.4956148 0.5059058
cg00327322_TC21 0.3515975 0.3787637 0.4452577
cg01203153_BC21 0.1617281 0.1447128 0.1514285

But after running NoMasked some values change ... Not all but some . I think there is some kind of re- arrangement that messes up the data.

test.betasCDPoMask[c("cgXXXX_BC11", "cgXXXY_TC21", "cgXXXZ_BC21"),][,1:3]
207222480102_R01C01 207222480102_R02C01 207222480102_R03C01
cg00054210_BC11 0.8903083 0.6598001 0.8574567
cg00327322_TC21 0.9022134 0.7865371 0.6779631
cg01203153_BC21 0.8391186 0.4313635 0.1078993

What I also realized is that, if a CpG-site has NAs for all samples It is not always dropped.. but rather somehow imputed!?

test.betasCDPo[7:8,]
207222480102_R01C01 207222480102_R02C01 207222480102_R03C01 207222480102_R04C01 207222480102_R05C01
cg00000289_TC21 NA NA NA NA NA

test.betasCDPoMask[7:8,]
207222480102_R01C01 207222480102_R02C01 207222480102_R03C01 207222480102_R04C01 207222480102_R05C01 207222480102_R06C01
cg00000289_TC21 0.8637285 0.832685 0.7981342 0.8453972 0.8275779 0.8382574

[zwdzwd]

I think it might have to do with the fact after getBetas(noMasked()), different samples are no longer of the same length. Could it be that cbind is causing the mislabeling of probe IDs?

[NazaninMir]

Im sorry, somehow my answer was not sent.
So here again:

I tried also the NoMasked with openSesame and I get also changed values.

I tried getBeta() on one sample with the and without the "NoMasked" function before and the values are changed.
Correctly the one ( cg00000289_TC21 ) with an NA is excluded, but the CpG-values are changed.

OneSample <- getBetas(idatlist_CDPo[[1]], mask=TRUE)
-->
cg00000158_BC21 cg00000165_TC21 cg00000221_BC21 cg00000236_TC21 cg00000289_TC21 cg00000292_BC21
0.9033180 0.2462405 0.8533618 0.7455211 NA 0.8637285
cg00000321_BC21
0.2538314

datlist_CDPo_ExcMask_open <- openSesame(idatlist_CDPo, func = noMasked,manifest=m2)
OneSample_ExcMask_open <- getBetas(idatlist_CDPo_ExcMask_open[[1]], mask=TRUE)

-->
cg00000158_BC21 cg00000165_TC21 cg00000221_BC21 cg00000236_TC21 cg00000292_BC21 cg00000321_BC21
0.9601634 0.1548068 0.9321163 0.8224250 0.9158312 0.2138720
cg00000363_TC21
0.3206771

[NazaninMir]

Ok here I tried NoMasked on one sample:
As you suggested it seems that using NoMasked with openSesame OR apply changes the structure of the list.

--> running both steps on one sample only dosen't change the values .. but it also do not exclude the NA

onesample <- noMasked(idatlist_CDPo[[1]])
onesampleBetas <- getBetas(exonesample, mask=TRUE)

cg00000158_BC21 cg00000165_TC21 cg00000221_BC21 cg00000236_TC21 cg00000289_TC21 cg00000292_BC21
0.9033180 0.2462405 0.8533618 0.7455211 NA 0.8637285
cg00000321_BC21
0.2538314