Inference of genetic regulatory network from microarray data is a fundamental objective of systems biology. I have gene expression profiles of some diseases with around 15,000 genes and 20 samples (microarray). I extracted differentially expressed genes using some of the existing statistical techniques. Now I am trying to find out the regulatory relationships among the extracted genes. Some of the techniques for the same are Entropy, MI, etc.
My question from the research community is how to find out regulatory relations using machine learning techniques (neural networks, genetic algorithm, etc.). Kindly explain in detail along with available software tools.
The "guilt by association" approach can be useful : starting from one of you most differientally expressed genes (or any gene that has a clearly relevant meaning in your study), you can compute (easily in excel) the correlation coefficient of all the other genes in your expression data, keep only the genes with the significant coefficients (that treshold depends on your number of samples), and then search for the enrichments in this list of transcriptionnal neighbours. For enrichments, a variety of easy-to use webtools are available : DAVID, GSEA,etc... my preference goes to the recently developped Enrichr, there : http://amp.pharm.mssm.edu/Enrichr/
and to its accompanying framework, L2N : http://amp.pharm.mssm.edu/lachmann/upload/login.php (where you can upload your list permanently, analyse the various enrichments and compute the sharing of transcriptionnal neighbours between lists). With the slight default that their graphical representation of networks needs to be improved still.. ;-)
In addition, don't forget that negative transcriptionnal neighbours (genes that correlate negatively but significantly with your gene of interest) are as interesting as positive neighbours.
Not quite sure what you want to do... If you only interested with interactome map (regulatory network) of the selected genes, you can create the map using Cytoscape (FREEware) -- a visualisation tool.
As far as I aware of there is no bespoke ANN or GA techniques that is FREE for use. you have to build it yourself (I mean coding)
Hi, depends of your skills in computational biology: if you are a computational biologist, or bioinformatician, I think this review M Bansal et al.,"how to infer gene networks from expression profiles",Molecular Systems Biology, 2007,is very useful. If you are an experimental biologist I think is better if you ask to an expert in computational: the use of these software require important knowledge in systems theory and statistics...otherwise they become a kind of black box in which you put the data but you don't know exactly how these data are processed by the software. Anyway, most of the computational methods, implemented in programming language and then called software, used to infer context specific regulatory networks are based on mutual information.
If you want to use a generic approach based on the information reported in literature and database,then not context specific, one of the most used software is Ingenuity (IPA), but it is no free.
Best
pas
A few R packages can be applied to this, e.g.
[1] BNArray: http://www.ncbi.nlm.nih.gov/pubmed/17005537
[2] AIRNet: http://www.biomedcentral.com/1471-2164/11/S2/S6
There are many options, you can compute correlations or other measures.
Always the causality inference, and how to discriminate between direct and indirect relations are big problems.
You can explore for instance the R package C3NET...
http://cran.r-project.org/web/packages/c3net/index.html
There are lot of inference algorithms which infer network (relations) between your genes based on the expression profiles.....you have to use their MATLAB, R codes as only some of them have GUIs ......GP dream is a good resource atleast for a few algorithms : have a look at this : http://dream.broadinstitute.org/gp/pages/index.jsf ....register a free account, prepare data as they suggested and run the algorithm.....and u can also read their papers for more details.....this is a very good paper on all network inference methods : http://www.ncbi.nlm.nih.gov/pubmed/22796662 ........
Hi Khalid,
It could be interesting to compare what you find with a large set of microarrays using the tool accessible here:
http://cardioserve.nantes.inserm.fr/madtools/home/
Go on the madcow section, which allow identification of genes that are coexpressed with a specified gene in multiple microarray data sets.
It permits experiment filtering and several levels of significance.
Good luck,
Pierre
The "guilt by association" approach can be useful : starting from one of you most differientally expressed genes (or any gene that has a clearly relevant meaning in your study), you can compute (easily in excel) the correlation coefficient of all the other genes in your expression data, keep only the genes with the significant coefficients (that treshold depends on your number of samples), and then search for the enrichments in this list of transcriptionnal neighbours. For enrichments, a variety of easy-to use webtools are available : DAVID, GSEA,etc... my preference goes to the recently developped Enrichr, there : http://amp.pharm.mssm.edu/Enrichr/
and to its accompanying framework, L2N : http://amp.pharm.mssm.edu/lachmann/upload/login.php (where you can upload your list permanently, analyse the various enrichments and compute the sharing of transcriptionnal neighbours between lists). With the slight default that their graphical representation of networks needs to be improved still.. ;-)
In addition, don't forget that negative transcriptionnal neighbours (genes that correlate negatively but significantly with your gene of interest) are as interesting as positive neighbours.
Dear researchers,
Thank you very much for your valuable inputs and insights. But my question is about "How to find out regulatory relations between genes" using SOFT COMPUTING TECHNIQUES"?
- Badges
- Science topic