Genetic Interaction Networks Inference

[Risclab]

Hello there,

 

its hard for me to grasp what is the difference between genetic interaction networks and gene regulatory networks?

I mean after all the interaction happens about the interaction via proteins?

 

Another question Im having is the inference of genetic interaction networks. What are the statistical tools for it

and more importantly what data is necessary for it? I mean genetical genomics is a prospering field, but is the

inference possible with expression data alone?

 

Thanks

R

[CharonY]

I think the terms are only very loosely defined and you will find these are similar terms pretty much describe the same thing, depending on whether you come from the biological or e.g. the bioinformatics side. They can be used to distinguish certain finer points depending on the way the network is built. For example the whole network can be based only on genes, i.e. you ignore the subsequent transcription and translation steps and potentially also involved metabolites. In this genes you would have genetic interactions mapped but you may not be able to read out the underlying mechanism of regulation. Or you can add protein information as well as level of regulation (which is more common when the term gene regulatory network is used).

But again, these tend only to be strict terms in a very small area in specific sub-disciplines. In common scientific usage one has to provide more context to explain what precisely is meant.

 

The question about inference is a fantastic question and I could write whole assays about that subject (which I should not due to time constraints). But to put it simple, it is one of the big challenges, how can genomic (or omics in general, i.e. transcriptomics, metabolomics and proteomics) information be used to inform us on physiological consequences?

The question is pretty much still open and there have been all kinds of approaches (often summarized under the banner of systems biology). However, there is still no clear consensus what the best way would be. Modeling techniques have seen some popularity and while useful for either simple networks for which we have a lot of biological information, they tend to be less useful for complex questions.

In the end, most of the time a clever experimental design is needed to really figure what is going on, with genomic information being just a part of it.

[Risclab]

Hello,

 

thank you for your answer. So when Im talking about gene network its just a gene regulatory network transferred to the gene space?

Lets consider for example the Biogrid database. It lists physical and genetical interactions. Whats the difference?

[CharonY]

I am not familiar with that database, but from the looks of it it appears that physical are include e.g. protein-protein or protein-DNA interaction (i.e. actual molecular interaction) whereas genetic interactions are apparently mostly based on mutational analyses or dose investigations (e.g. effects of over- underexpression).

I expect that in some cases there will be considerable overlap.

[Risclab]

Ah so the genetic interaction is more like an functional relationship, where I ignore the protein interactions and just look at the effect of the genes in terms of the phenotype?

[CharonY]

In principle yes. I think for this database it is decided on the type of experimental evidence (e.g. using mutant analyses vs DNA binding assays).

[Risclab]

Ah ok. Its basically an abstraction into the gene space? So thats the same when I consider epistatic interactions?

 

So what are the statistical methods to infer these gene interaction networks? Can I infer these networks from expression

data alone?

[CharonY]

Depends really on the experimental setup rather than statistical methods. I believe the database you linked to explicitly used literature data rather than using raw data and infer relationships from there. There are statistical methods to model and look at network relationship but they are a bit hit and miss.

Most often a standard experiment would look like this:

Mutate potential regulator, check whether putative target is up or down regulated, (using statistics to compare to wildtype). As the expresion data is relative (usually) it is almost impossible to infer anything outside of the context of the experimental setup.

 

In other words, if you only had raw expression data you generally would not be able to reconstruct how the elements interact with each other.

 

Edit: I should add that it obviously also depends on the type of expression data. One could for example apply multivariate statistics to infer interactions. However, it is usually not terribly accurate.

Edited October 4, 2013 by CharonY

[Risclab]

Ah I see. One last question for the case of gene interactions as specified in BioGRID where mutliple gene variants determine the outcome

of a specific phenotype. Whats the general approach for these. I mean conducting epistasis analysis or exploiting naturally occuring perturbation

(eQTL data) is statistical framwork is an oppurtunity to infer these. However, how about simple gene expression analysis or gene perturbation screens?

It is possible to infer these interactions without exact genetic information? And how about statistical methods for them? Boolean networks, Bayesian networks,

SEM. What are the current methodologies?