[R] how to calculate multiple meta p values

[Ana Marija]

Hi Michael,

this still doesn't work, by data frame has a few less columns now, but
the principle is still the same:

> head(d)
    chr    pos         gene_id                     LCL
Retina       wl           wr
1: chr1 775930 ENSG00000237094 0.3559520 9.72251e-05 31.62278 21.2838
2: chr1 815963 ENSG00000237094 0.2648080 3.85837e-06 31.62278 21.2838
3: chr1 816376 ENSG00000237094 0.3313120 3.85824e-06 31.62278 21.2838
4: chr1 817186 ENSG00000237094 0.0912854 3.75134e-06 31.62278 21.2838
5: chr1 817341 ENSG00000237094 0.1020520 3.75134e-06 31.62278 21.2838
6: chr1 817514 ENSG00000237094 0.0831412 3.82866e-06 31.62278 21.2838

so solution for the first row should be:
> sumz(c(0.3559520,9.72251e-05), weights = c(31.62278,21.2838), na.action = na.fail)
sumz =  2.386896 p =  0.008495647

when I run what you proposed in the last email:

helper <- function(x) {
  p <- sumz(as.numeric(x[4:5]), weights = as.numeric(x[6:7]))$p
  p
}

d$META <- apply(d, MARGIN = 1, helper)

I am getting:

Error in sumz(as.numeric(x[4:5]), weights = as.numeric(x[6:7])) :
  Must have at least two valid p values

Please advise,
Ana

On Wed, Oct 30, 2019 at 5:02 AM Michael Dewey <lists using dewey.myzen.co.uk> wrote:
>
> Dear Ana
>
> Yes, when apply coerces q to a matrix it does so as a character matrix
> because of the values in the first column. So you need to wrap the
> references to x in helper in as.numeric() tat is to day like
> as.numeric(x[2:4]) and similarly for the other one. Sorry about that, I
> should have thought of it before.
>
> When I next update metap I will try to get it to degrade more gracefully
> when it finds an error.
>
> Michael
>
> On 28/10/2019 19:06, Ana Marija wrote:
> > Hi Michael,
> >
> > I tried what you proposed with my data frame q:
> >
> >> head(q)
> >             ID                P             G              E
> >   wb          wg           we
> > 1:  rs1029830 0.0979931 0.0054060 0.39160 580.6436 40.6325 35.39774
> > 2:  rs1029832 0.1501820 0.0028140 0.39320 580.6436 40.6325 35.39774
> > 3: rs11078374 0.1701250 0.0009805 0.49730 580.6436 40.6325 35.39774
> > 4:  rs1124961 0.1710150 0.7252000 0.05737 580.6436 40.6325 35.39774
> > 5:  rs1135237 0.1493650 0.6851000 0.06354 580.6436 40.6325 35.39774
> > 6: rs11867934 0.0757972 0.0006140 0.00327 580.6436 40.6325 35.39774
> >
> > so the solution of the first row would be this:
> >> sumz(c(0.0979931,0.0054060,0.39160), weights = c(580.6436,40.6325,35.39774), na.action = na.fail)
> > sumz =  1.481833 p =  0.06919239
> >
> > I tried applying the function you wrote:
> > helper <- function(x) {
> >    p <- sumz(x[2:4], weights = x[5:7])$p
> >    p
> > }
> >
> > With:
> >
> > q$META <- apply(q, MARGIN = 1, helper)
> >
> > # I want to make a new column in q named META with results
> > but I got this error:
> >   Error in sumz(x[2:4], weights = x[5:7]) :
> >    Must have at least two valid p values
> >
> > Please advise,
> > Ana
> >
> > On Sun, Oct 27, 2019 at 9:49 AM Michael Dewey <lists using dewey.myzen.co.uk> wrote:
> >>
> >> Dear Ana
> >>
> >> There must be several ways of doing this but see below for an idea with
> >> comments in-line.
> >>
> >> On 26/10/2019 00:31, Ana Marija wrote:
> >>> Hello,
> >>>
> >>> I would like to use this package metap
> >>> to calculate multiple o values
> >>>
> >>> I have my data frame with 3 p values
> >>>> head(tt)
> >>>             RS            G           E          B
> >>> 1: rs2089177   0.9986   0.7153   0.604716
> >>> 2: rs4360974   0.9738   0.7838   0.430228
> >>> 3: rs6502526   0.9744   0.7839   0.429160
> >>> 4: rs8069906   0.7184   0.4918   0.521452
> >>> 5: rs9905280   0.7205   0.4861   0.465758
> >>> 6: rs4313843   0.9804   0.8522   0.474313
> >>>
> >>> and data frame with corresponding weights for each of the p values
> >>> from the tt data frame
> >>>
> >>>> head(df)
> >>>          wg       we             wb                RS
> >>> 1 40.6325 35.39774 580.6436 rs2089177
> >>> 2 40.6325 35.39774 580.6436 rs4360974
> >>> 3 40.6325 35.39774 580.6436 rs6502526
> >>> 4 40.6325 35.39774 580.6436 rs8069906
> >>> 5 40.6325 35.39774 580.6436 rs9905280
> >>> 6 40.6325 35.39774 580.6436 rs4313843
> >>>
> >>> RS column is the same in df and tt
> >>>
> >>
> >> So you can create a new data-frame with merge()
> >>
> >> newdata <- merge(tt, df)
> >>
> >> which will use RS as the key to merge them on.
> >>
> >> The write a function of one argument, a seven element vector, which
> >> picks out the p-values and the weights and feeds them to sumz().
> >> Something like
> >>
> >> helper <- function(x) {
> >>    p <- sumz(x[2:4], weights = x[5:7])$p
> >>    p
> >> }
> >> Note you need to check that 2:4 and 5:7 are actually where they are in
> >> the row of newdat.
> >>
> >> Then use apply() to apply that to the rows of newdat.
> >>
> >> I have not tested any of this but the general idea should be OK even if
> >> the details are wrong.
> >>
> >> Michael
> >>
> >>
> >>> How to use this sunz() function to create a new data frame which would
> >>> look the same as tt only it would have additional column, say named
> >>> "META" which has calculated meta p values for each row
> >>>
> >>> This i s example of how much would be p value in the first row:
> >>>
> >>>> sumz(c(0.9986,0.7153,0.604716), weights = c(40.6325,35.39774,580.6436), na.action = na.fail)
> >>> p =  0.6940048
> >>>
> >>> Thanks
> >>> Ana
> >>>
> >>> ______________________________________________
> >>> R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
> >>> https://stat.ethz.ch/mailman/listinfo/r-help
> >>> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> >>> and provide commented, minimal, self-contained, reproducible code.
> >>>
> >>
> >> --
> >> Michael
> >> http://www.dewey.myzen.co.uk/home.html
> >
>
> --
> Michael
> http://www.dewey.myzen.co.uk/home.html