Entering edit mode
7.0 years ago
lur_murad
•
0
excuse my post I am in pretty new in Bioinformatics field. I analysed PPI network after integrated gene expression data from T2D (Type 2 diabetes) experiment within PPI network and reveals some sub network. First, I used (limma package) for Differentially Expressed Gene analysis. Second, I mapped DEG genes on the PPI network and assign the gene fold change value to corresponding proteins. Third, I search the network by selected my candidate gene and reveals sub-networks. I scored them by my formula, then I merge the top scoring sub networks. Now, I want to validate my results (merged sub network) and I have no idea how to do
Could anyone help me or suggested a method to validate my outcome please? I will highly appreciated
1) Gene Ontology 2) Experimental Validation
@kennethcondon2007 Thank you for your reply. How can I do that please?
I saw some researcher use k-fold cross validation. However, I did not separate the samples to test and training when I mapped it on the network.
What you want to validate is not entirely clear. Cross-validation is used when training supervised machine learning models which is not something you seem to be doing. Assuming that what you want to validate are computationally-derived conclusions or predictions, @kennethcondon2007 suggested two commonly-used approaches:
1- to use Gene Ontology annotations to test for enrichment in particular terms in your sub-networks (or any partition of your final data that is relevant) and see if they are consistent with your expectations.
2- do experiments to validate your computationally-derived predictions. However, it seems that you may be interested in some independent confirmation of your final network. Depending on what kind of relation(s) the edges represent, the best way is probably to experimentally test them. Alternatively (or complementary to experimental validation), you could check whether these relations already exists in relevant public databases.
I did the following steps : 1. I used (limma package) to find Differentially Expressed Gene for T2D (Type 2 diabetes) microarray. 2. I import PPI interaction networks from IntAct repositories. 3. I upload my list of genes with their expression/fold change values and map them onto the network. 4. I used Depth-first search to identify sub-networks of genes/proteins which are highly expressed and connected within the network 5. I scored these sub-networks by using a formula and then filtered them to get the most significant and highly scored ones. 6. I visualize the results. and now I want to valiadte my results to see if this results relevants to T2D and if I can use it as a T2D biomarker
I don't get what is supposed to be the biomarker. Is it the whole selected sub-network i.e. including the interactions or is each one of the genes in this sub-network a potential biomarker ? In the case of a biomarker, validation usually means that the measurements made of the marker correlate with some clinical outcome. Protein interactions usually don't make good biomarkers because they are not easy to measure in cells or tissues. You can get an idea of the relevance to T2D by checking if the final gene list is enriched in genes known to be involved in T2D or T2D-affected processes.
@jean-Karim Heriche No the whole subnetwork (gene and their interactions)- network based biomarker to distinguish between disease and normal state. I got most of the gene in my sub-network involved to T2D when I used GO I am not sure how to validate that