SMLG (Statistical Machine Learning Group) Discussion Forum

I took Dr. Yoo's advice and created a data set using the top 50 positively correlated genes and the top 50 negatively correlated genes. The CCGrade100Genes.csv contains this data set. The CaseControlGrade100Genes.xlsx shows how I determined the genes to use. In the zip file, it also includes the code creating the networks. It uses the Max-Min Hill Climbing Method, Grow-Shrink method, and the Hill Climbing Method. It also shows the arcs and the overall network for each algorithms. It also includes code to perform a 10-fold cross validation for each network. The grow-shrink and max-min hill climbing algorithm do not create a robust network but the hill climbing one does. I saved the images of the created networks as PDFs. In addition, I started experimenting with using banjo. I attached the settings file for using the greedy hill climbing algorithm. Changes need to be made to the code if you want to run it on your machine.
1. Download banjo and in the data folder create a folder called project then an input and output folder. data/project/input and data/project/output.
2. Put the CCGrade100Genes.txt file in the input folder.
3. Change the time accordingly to how long you want it to go. Change maxTime = 60 m to something else like maxTime = 120 m etc
4. Install Graphviz and put the path for where the dot.exe is located. fullPathToDotExecutable = C:/Program Files (x86)/Graphviz2.38/bin/dot.exe
5. Open CMD and go to the folder containing banjo.jar
6. Enter java -jar banjo.jar settingsFile=data/project/CCGBNsettings.txt
After that the code will run and in the output file, it will give a report as well as a picture of the network if you include the correct path of Graphviz.

Here is the project. Everything used is in the zip file as well as the paper.