I'm developing and testing some algorithms that scores drug combinations based on how they disrupt the topology of a network. To be able to compare results and look for correlation I will compare my algorithms with data from the NCI Almanac that has data on approximately 5000 drug combinations on the 60 cell lines in NCI-60. The problem is I'm not 100% sure how to create the networks that would represent a specific tumor cell line. I have a couple of articles that has shown correlation in cancer 5-y survival and the degree of complexity that I want to test the same methods to see if I can get similar results. the articles are: Degree-entropy and persistent homology, but they dont explain how the networks was constructed.
I've used a lot of time trying to figure this out where I ended up creating a ppi network selecting the most expressed genes (top 20%) then building a network using iRefIndex in R, but I only have 3 samples for each cell line from here. I know CellMiner has a lot of data on the NCI-60, but I don't know how to fully utilize it and would love to get some pointers from you guys. I even know there exists an R package to access CellMiner so maybe its best to use this? Would the solution be to use a mix of samples from different platforms (Affymetrix U133 Plus 2.0, RNA: Affy HG-U95(A-E), ... ) and somehow retrieve the most expressed genes from them, or something completely different?
I appreciate all help I can get. Then I can finally run the application I have created on correctly generated input networks so I can validate the algorithms.
...but, is low expressing not also important to consider? Why did you just pick the most expressed genes?
Creating an individual network on just 3 samples is not great but, if that's all that you can get for each cell line, then you could at least make a start. Merging data across different platforms is definitely not the best approach; however, if you normalise each independently and then convert the normalised values to Z-scores (separately for each platform), then I believe you have a better chance of merging them and creating a network from the Z-scores.