Dear all friends,
I am writing to ask you about ARACNE plugin in Cytoscape. Is there anyone who has worked with this plugin? I am trying to reconstruct GRN based on expression data using the ARACNE algorithm. Although ARACNE handle low number of sample, the interaction number decrease when the sample number increase. For example for a dataset with 9000 genes and 18 samples, it reconstruct a GRN contains 7000 interactions; on the other hand, it reconstruct a GRN contains 900 interactions when the sample size is 36.
For your information, I tried different algorithm modes and different DPI. Finally, I set the “Discover” mode for the algorithm and “Adaptive Partitioning” for MI Algorithm Type, DPI = 0.15, MI threshold = 0.5. I also added a list of human transcription factors as hub and TFs.
Because the comparison of the network diameter in cancer and normal networks is important for me, I decided to select the same sample size for both normal and cancer condition. For solving the problem:
I used hierarchical clustering method, first. I clustered 36 cancer samples into two groups and make the separate GRNs using them. I hoped to get close result, however the result was surprising: It shows completely different GRNs for two networks that both are constructed by Breast tumor samples even from the same platform and the same datasets (just 10 common interactions between 5000 interactions, for example).
I also tried to expand the normal sample size using merging normal expression data from different datasets but it is not suitable for the project, because we need the exact same condition, for example the fact that data obtained from which tissue, stroma or epithelial tissue, is important. Besides, merging data is difficult and noisy.
Now I decide to generate data for normal condition based on the present data using different statistical methods.
I wonder if there is anyone with better idea to solve the problem.
Look forward to hearing from you.