Pulmonary Arterial Hypertension - PAH

[Hamza]

Dear all,
Here is my current dataset for my research on PAH. It still need some modifications in order to analyze it.

[cwyoo]

Use random number gerator, e.g., http://www.random.org/sequences/, generate random number sequences. For example, let say you got the following sequences (smallest 1 and largest 10):
2
5
3
8
10
...

For the first repeating measurements (assume there are three repetitions), you use the first random number 2, i.e., select the second measurement for that gene. For the second repeating measurements (assume there are two repetitions), you use the first random number 5, i.e., select the first measurement for that gene.

When do you think you can complete the dataset? Once completed, please contact Luis for Bayesian networks analysis with banjo.

[Hamza]

I randomly selected only duplicate expressions in some studies since there was only one expression in most studies.
Also, I transposed the genes to be in columns and samples to be in rows.

Please let me know for any comment on the attached file (see the last two sheets).

[Hamza]

Here is the final file for PAH. I descritized the data again using this formula:
if ((value <mean-S.E,0,if(value>mean+S.E,2,1)) for each study.

Yesterday, Luis and I applied the data file on Banjo for 1, 2, and 4 hours, but we didn't get the same results among 3 maps. We run it for 6 hours and will check the results tomorrow.

For SPSS, I want to make sure that I will use PAH column as dependent factor and the genes as independent?

[cwyoo]

Yes, for logistic regression, you should use the original data (before discritization). Also please post here the settings and results (images of only the top networks) of the different runs.

[meninonas]

Dr. Yoo,

I just checked with Hamza the 6-hour run and none of the three network matched, which means that neither the 1, 2, 4, nor 6 hour run created equal BNs.

I was thinking that maybe if Hamza runs the logistic regression first and then finds which genes are associate with Hypertension, we could run Banjo with just these genes. I think that it makes more sense when it comes to the creation of the Bayesian Network and also it might help when we run Banjo with less variables.

Lastly, I was wondering about Logistic vs. Linear Regression for his dataset. We found that linear Regression works when both the dependent and independent variables are continuous. In Hamza's dataset, PAH (which is the dependent variable) is binary; hence, linear regression doesn't seem to work very well. In contrast, Logistic Regression allows for one of the variables to be categorical. Do you think we should try to use both or just logistic regression?

Websites where I found information about Linear Regression:

http://cjem-online.ca/v9/n2/p111
http://www.adasis-oz.com/tips/2013/8/28 ... regression
http://udel.edu/~mcdonald/statlogistic.html
http://stackoverflow.com/questions/1214 ... regression

[Hamza]

Here is 1 hour run for banjo, they are 3 networks. and I will post 2, 4, and 6 hours separately.

[Hamza]

This is 2 hours run networks.

[Hamza]

This is 4 hours run networks

[Hamza]

Lastly, the 6 hours run networks. None of the run had the same networks.
As Luis and I were discussing, 6 hour in the maximum time to run by Banjo. However, as Luis mentioned, it might be helpful if I did logistic regression in SPSS and then use the significant genes which will be less in number and run them in Banjo. What is your suggestion?

I am still working on the best setting for logistic regression in SPSS since I had some issues with it today.