Tutorial - 4 - Setting Parameters

Index - 1  - 2 - 3 - 4 - 5 - 6 - 7<previousnext>

Here is a screenshot of the Analysis section: through this panel, the user can select the data file

 

 

and set the parameters that will be used in the analysis.

 

For example we suggest to use Legendre as basis function, maximum degree L of polynomial to be about half of the (distinct) time points available, to not provide smoothness information and hence set vi=0

 

 

 

To choose the parameter lambda of the truncated Poisson to match the expected prior degree, which is reasonable to lay between L/2 and L-1.

 

Bayesian Analysis for Time Series

Microarray Experiments

Casella di testo: Then any user can choose one of the three prior models contained in BATS.
For the prior model 2, BATS offers the user the option either to choose one parameter of the distribution and to estimate the second using the method of moments (Choice 1) or to let the software estimate both of them automatically from the data using MLE estimators (Choice 2).

Casella di testo: The user may then choose to print out the estimated profiles (and the raw data ) for the top “nfirst” genes in the ranked ordered list
 If Global scale is selected, then all genes will be depicted on the same axis (and thence the figure will be comparable), if Auto scale is selected, then each gene will be depicted on the most appropriate axis (and then it will be easier to examine each profile.)

Here the user can select estimator for the global parameters     

(note that user specified choices are allowed)

and chose the method for estimating the gene specific degree

(two options are avaliable, MAP and Mean)

And the range for maximizing the marginal probability distribution function of the data and hence estimate 

The user may control the Multiplicity error using the procedure proposed in  Abramovich & Angelini, Bayesian MAP multiple testing procedures Sankhya (2006).

 

Binomial prior is usually found to be the most appropriate.

 

However, with the option Rank ordered is also possible to print out the ordered list of genes without automatically determine the cut-off of significance. The biologist can after decide to select only the top gene in the list without carrying out any formal test of significance.

A user friendly software for Bayesian Analysis of Time Series Microarray Experiments.

For a short description of the statistical model implemented in BATS see here 

Elaborazione alternativa: BATS-MODELS
Casella di testo: Choosing appropriate parameters for an analysis on a particular data-set with BATS usually requires some preliminary knowledge of statistics and some level of expertise.  A not expert user should not be discouraged, since for all parameters  a default value is provided. We observe that some of the default values are directly computed when the dataset is loaded.  We believe that in most of the cases the default values can be used and for those user how are not sure on how to modify them we suggest to leave the such values.
Moreover from our study we found that the biological findings were robust with respect the choice of the parameters, provided they are chosen in a reasonable range. 
In the following we provide detailed explanation on how to choose the parameters.