[BioC] RMA background adjustment implementation in affy

[Mikko Korpela]

Hello!

There are some differences in the way the RMA background adjustment
formula is presented in different documents, and the way it's implemented
in the affy package. I wonder what causes the differences, and whether I
should be worried about the RMA values produced by affy.

The relevant part of the code, I think, is this (copy-paste from affy
1.9.6, same as in many previous versions of affy, at least in affy-1.5.8):

---

void bg_adjust(double *PM,double *MM, double *param, int rows, int cols, 
int co$  int i;
  double a;

  for (i=0; i < rows; i++){
    a = PM[column*rows + i] - param[1] - param[0]*param[2]*param[2];
    PM[column*rows + i] = a + param[2] * phi(a/param[2])/Phi(a/param[2]);
  }

}

---

where the array "param" is as described by the following comment (also
copy-paste):

 ** note we will assume that param[0] is alpha, param[1] is mu, param[2]
is sigma

The documents that do not agree with the above are builtinMethods.pdf
(documentation of the affy package) and
http://stat-www.berkeley.edu/users/bolstad/stuff/AffyLowLevelWorkshop_Poster_20$

In particular, some terms are missing from both the numerator and the
denominator of the division operation in the affy implementation. The
documents also do not agree with each other...

Am I missing something simple here? I hope someone knows the right
formula for the background adjustment and the reason why the
implementation is different (it seems) from the documentation.

Thanks,

Mikko