Epistasis Blog

From the Artificial Intelligence Innovation Lab at Cedars-Sinai Medical Center (www.epistasis.org)

Friday, May 20, 2005

MDR 0.3 released

The Dartmouth Computational Genetics Laboratory (CGL) is pleased to announce the release of version 0.3 of our multifactor dimensionality reduction (MDR) software for detecting and characterizing gene-gene and gene-environment interactions in genetic and epidemiologic studies of common human diseases.

The new version of MDR can be downloaded from here.

New features include:

1) Publication-quality figures of the multilocus MDR models. Models can be viewed in any dimension and saved to an encapsulated postscript (*.eps) file.

2) The ability to save a snapshot of an MDR analysis so you can load it later. This eliminates the need to rerun an MDR analysis if you forget to output a set of results. All aspects of the analysis are saved including the data, the configuration parameters used, and all the results.

3) A sign test to test the null hypothesis that number of testing accuracies with values > 0.5 (+)is equal to the number <= 0.5 (-) from 10 or 20 cross-validation intervals, for example. We think the sign test might be useful for deciding whether or not to proceed with a computationally expensive permutation test using the MDR Permutation Testing module. We don't see it being used for formal hypothesis testing unless the size and power of the test are determined to be acceptable.

4) The ability to save the raw results from the MDR analysis. Raw results now include all fitness values (average training accuracies) for every model considered.

5) The ability to consider matched case-control or family-based data. Here, matched pairs are kept together during cross-validation.

6) IF-THEN rules for each genotype combination and their "high-risk" or "low-risk" assignments. For example: IF SNP1 = AA and SNP2 = GG then 1.

7) A progress bar to determine how long of a lunch break to take.

8) A new data format in which the class variable is the last column and each column has a header describing that variable or attribute (e.g. SNP1, CYP2D6). Old MDR data formats can be converted to new MDR formats using the MDR Data Tool availabe here.

Here are a few features that are planned for future releases:

1) Threading to take advantage of multi-processor computers.

2) Batch/command line mode to allow MDR to be run from scripts. This will facilitate running MDR multiple times in simulations on a parallel computer.

3) Visualization of the fitness landscape.

4) Wrapper-algorithms for variable/attribute selection. This will be important when the number of attributes is too large for exhaustive searching.

5) A context-sensitive help system.

Is there something you would like to see added to MDR? Request it here.

Note that MDR will be in beta testing for another 2-3 months. Please send us your feedback so we can roll out a polished MDR 1.0 later this summer.

0 Comments:

Post a Comment

<< Home