LMM software tool is a Java implementation of the light mutual min (LMM) algorithm which was proposed as a fast and robust constraint based learning of Bayesian networks (BN) from observational data. The tool provides implementation for both, learning the skeleton of the Bayesian network and the orientation of the edges in the skeleton to recover the completed partially directed acyclic graph (CPDAG). The tool is developed with sufficient flexibility to easily allow users to incorporate their own conditional independence methods. Also, other researchers will be able to use either of the skeleton recovery or the orientation of the edges with a third party method. For example, researchers, who work on skeleton recovery methods, can use the edges orientation part of LMM to orient their skeleton into CPDAG or vice-versa.
- LMMTool was developed and tested using Java version 6.
- LMMTool supports both the recovery of the skeleton and the recovery of the completed partially directed acyclic graph (CPDAG) from a given skeleton.
Current version supports the partial correlation for conditional independence which can be used with multivariate continuous data.
- Following the guidelines presented in the enclosed manual, users should be able to develop and incorporate new conditional independence testing methods.
- Java version 6 or higher needs to be installed to be able to compile and run LMMTool.
- Non-expert java users are advised to download Java development environment such as JDeveloper (http://www.oracle.com).
Dependence on external packages: The following packages has to be available in the classpath:
- Commons-math package: http://commons.apache.org/math/.
- Jama package: http://math.nist.gov/javanumerics/jama/.
- The full paper of our methods and experimental results can be access at: http://jmlr.org/papers/v14/mahdi13a.html.
The zip file of our tool contains:
- Full source code.
- Manual of how to use and customize LMMTool.
- Supplementary document of our main paper that contains further discussion of the main methods.
A modified version of LMM was presented and tested in  to construct gene interaction networks from gene expression data. The implementation in  relies heavily on empirical estimation of independence testing parameters and takes in consideration issues related to the nature of gene expression data and gene interaction networks. The empirical LMM was shown to be competitive to the state of the art methods in gene networks recovery and in many cases had the best performance.
 - R. Mahdi and J. Mezey, “Sub-Local Constraint-Based Learning of Bayesian Networks Using A Joint Dependence Criterion”, Journal of Machine Learning Research, vol. 14, pp. 1563-1603, 2013.
 - R. Mahdi, A. S. Madduri, G. Wang et al., “Empirical Bayes conditional independence graphs for regulatory network recovery,” Bioinformatics, vol. 28, no. 15, pp. 2029–2036, 2012.
- Changes to previous version:
Initial Announcement on mloss.org.
No one has posted any comments yet. Perhaps you'd like to be the first?
Leave a comment
You must be logged in to post comments.