[Biopython] quantile normalization method

[Laurent Gautier]

Hi Bartek and Vincent,

Few comments:

A/

The algorithm is fairly straightforward, as you noted it, but beware of 
details such missing values, ability to normalize against a target 
distribution, or ties when ranking (although I'd have to check if those 
receive a special treatment).
The quantile normalization code in the R package "preprocessCore" is in 
C and might outperform a pure Python implementation.

B/

There is a variety of normalization methods in bioconductor, and it 
might make sense to embrace it as a dependency (rather than reimplement 
it). I have bindings for Bioconductor up my sleeve about to be 
distributed to few people for testing. The public release might be 
around ISMB, BOSC time.

C/

norm_a = numpy.array(normq(m))

can be replaced by

norm_a = numpy.as_array(normq(m))

to improve performances whenever m is of substantial size (as no copy is 
made - see 
http://rpy.sourceforge.net/rpy2/doc-2.1/html/numpy.html#from-rpy2-to-numpy )



Best,


Laurent



On 3/20/10 5:00 PM, biopython-request at lists.open-bio.org wrote:
>> >  Is there a quantile normalization method in biopython, I search but did not
>> >  find. If not it looks straight forward would it be of any interest to the
>> >  community for me to contribute a method
>> >
>> >  1. given n arrays of length p, form X of dimension
>> >  p ? n where each array is a column;
>> >  2. sort each column of X to give X sort ;
>> >  3. take the means across rows of X sort and assign this
>> >  mean to each element in the row to get X sort ;
>> >  4. get X normalized by rearranging each column of
>> >  X sort to have the same ordering as original X
>> >
>> >  From
>> >  A comparison of normalization methods for high
>> >  density oligonucleotide array data based on
>> >  variance and bias
>> >  B. M. Bolstad 1,?, R. A. Irizarry 2, M. Astrand 3 and T. P. Speed 4, 5
>> >  ?
>> >
>   Hi,
>
> I don't think there is such a method available.
>
> I'm myself using the original R implementation by Bolstad et al. It requires
> rPy and R installed. It can be achieved in a few lines of code:
>
> <pre>
> import rpy2.robjects as robjects
> #ll = list of concatenated values to normalize
> v = robjects.FloatVector(ll)
> #numrows=number of vectors that made up ll
> m = robjects.r['matrix'](v, nrow = numrows, byrow=True)
> robjects.r('require("preprocessCore")')
> normq=robjects.r('normalize.quantiles')
> norm_a=numpy.array(normq(m))
> #norm_a=normalized array
>   </pre>
>
> If your method is a pure python implementation which is comparably fast I
> think it would be worth to have it in Biopython since the method is (in my
> opinion) quite useful and it would remove the dependency on R from some of
> my scripts.
>
> cheers
>   Bartek
>