[BioC] toptable with multiple coefs [was Time course experiment....]

[Gordon K Smyth]

On Tue, January 3, 2006 11:24 pm, Sean Davis wrote:
> On 1/2/06 8:00 PM, "Gordon Smyth" <smyth at wehi.edu.au> wrote:
>> topTable() doesn't have a facility for sorting on or presenting the
>> F-statistics, because it is individual coefficient orientated.
>>
>> I have toyed with the idea that perhaps topTable() should output a
>> table based on the F-statistic when the argument 'coefficient' is set
>> to NULL. In other words topTable() would give individual coef
>> significance when a coef is specified, otherwise it would give
>> overall significance. Would you find that a useful facility?
>
> Gordon,
>
> I do think your suggested additions would be useful.  In addition, I often
> find myself generating a large spreadsheet based on the output of a number
> of individual coefficients so that the biologist can use that spreadsheet in
> something like Excel to sort and filter data at will.
>
> While less statistically palatable than the current method and your F-stat
> extension, would it be possible to include a third variation such that the
> coefficient argument could be set to a vector?  This then leaves the
> question of how to "rank" or "sort" genes, but one could choose arbitrarily
> to order by the input to topTable (the order of the fit object--I typically
> use ALL genes for output) or to use the first value in the coefficient
> vector as the "key" coefficient.  For complicated experiments, I think many
> biologists like to see how a gene looks for multiple coefficients
> simultaneously.
>
> Thanks,
> Sean

Hi Sean,

I'm not sure what output you are thinking of for vector coef.  How is it different from

  write.fit(fit)

or

  write.fit(fit[,selectedcoefs])

?

In general, the use of contrasts is intended for looking at subsets of coefficients.  E.g.,
contrasts.fit() followed by topTable() with coef=NULL could rank genes on the F-statistic for any
selection of coefficients or contrasts.

My thought of what a vector coef argument to topTable() would be is that it should (i) compute the
fit object for the set of contrasts defined by those coefficients and (ii) output a top-table
ranked by the F-statistics for that set of contrasts.

Cheers
Gordon