Newly re-written version of hdarray.
The bayesreg code
performs the
same calculations as hdarray except for a few usage and interface related differences. The
differences between the hdarray code and bayesreg code are:
Usage: bayesreg is mainly intended for people who want to use the R environment (and not the Cyber-T web
interface) to
run their data. bayesreg returns one large matrix containing the computed statistics on the data, including the ROC x
and y-coordinates and PPDE values. hdarray does not return a matrix, it creates a few output files (Allgenes, mix.txt
etc)
Interface: The user interface is simpler, the arguments are fewer (see the "readme" file on the download
page
after
registration). The user needs to ensure that the dataset contains all the experiment measures in order (i.e. all
control columns first and then experimental columns) and no additional columns. Then the user just needs to invoke
bayesT(myData,numControlColumns,numExptColumns). Similarly, for the paired t-test, the function is
bayesT.pair(myData,numRatioColumns). Optional arguments include the window size, conf etc., as in hdarray.
Minimum replicates: If a gene in any condition has fewer than 2 replicates, the gene will be
ignored in the t-test and the t-statistics will output "NA".
Calculation: Hdarray returns statistics on the input data and log-transformed-input data. This sort of
implicitly assumes that the data input are raw or normalized non-log-transformed intensities. bayesreg does not
perform any internal log-transformations on the data (except for the bayesT.pair, if the option is set). All data are
assumed to be
transformed (normalized etc.), if needed, prior to invoking the bayesT function. Thus the input to bayesreg can be
already log-transformed data, for example.
Missing data: Missing values are to be assigned as "NA" by the user. Zeros are considered valid data. For
example, if
the
raw data contained "1" and the dataset is log-transformed, you will have "0" in the data. This will be considered as a
valid replicate datapoint. If you want the program to ignore "0", then you need to assign them as "NA". hdarray ignores
both "0" and "NA", bayesreg ignores only "NA".
bayesAnova
One way ANOVA with bayes regularization of standard deviation
(extension of the bayes T-test to multiple
conditions). Note: Beta version currently available.
For bug reports and comments,
please send us e-mail.
This R program performs one-way ANOVA on replicate measurements of more than two conditions. See the "readme" file
on the download page following registration for details on the usage and parameters.
Post-hoc pairwise comparison functions are also part of this code. Two methods - TukeyHSD and Scheffe's test
for comparison of all pairs of means have been implemented.
Minimum replicates: If a gene in any condition has fewer than 2 replicates, that condition will be
ignored in the ANOVA F-test (for that particular gene only) and correspondingly in the postHoc test.
Calculation: Hdarray returns statistics on the input data and log-transformed-input data. This sort of
implicitly assumes that the data input are raw or normalized non-log-transformed intensities. bayesreg does not
perform any internal log-transformations on the data (except for the bayesT.pair, if the option is set).. All data are
assumed to be
transformed (normalized etc.), if needed, prior to invoking the bayesT function. Thus, the input to bayesreg can be
already log-transformed data, for example.
Missing data: Missing values are to be assigned as "NA" by the user. Zeros are considered valid data. For
example, if
the
raw data contained "1" and the dataset is log-transformed, you will have "0" in the data. This will be considered as a
valid replicate datapoint. If you want the program to ignore "0", then you need to assign them as "NA". hdarray ignores
both "0" and "NA", bayesreg ignores only "NA".