Hello,
We have confirmed that all algorithm steps involved in both RMA and
GC-RMA are identical between the Bioconductor packages and ArrayAssist
3.3. As some people have suggested before in this thread, the
differences in outcome arise solely as a result of implementation
choices--sometimes Bioconductor packages will employ routines coded in
R, where ArrayAssist uses high-performance C++ libraries and native
C/C++ and Fortran code.
We have performed extensive comparative analysis of all final and
partial results from Bioconductor and ArrayAssist implementations of
the
algorithms in question. The analysis confirmed that the issues are
limited to implementation differences. Algorithm steps coded using the
same language produced identical results in both environments.
Algorithm
steps coded using different languages (and converted to executable
code
using different compilers) produced slightly different results.
Practically, the differences are limited to the calculation of loess
coefficients for background estimation.
We are consistently obtaining correlation values between Bioconductor
and ArrayAssist in excess of 0.99, often better than 0.9999. Please
note
that different data sets will produce different results due to varying
data distribution and to the probabilistic component involved in the
calculations.
Bogdan Georgescu
Stratagene Software Solutions
PS. Philip, the latest version of ArrayAssist may be downloaded using
the Check for Updates feature. The corrections you mention are all in
version 3.3.
