A key component of my PhD thesis -all those years ago- was called Nested Effects Models (NEMs) and I feel lucky and privileged that over the years many other researchers have liked them enough to apply them in their work and extend the methodology. Before I show you two recent examples, here is the 5 minute summary of how NEMs work:
What are Nested Effects Models?
NEMs were introduced in 2005 in a paper titled “Non-transcriptional pathway features reconstructed from secondary effects of RNA interference.” The idea is simple: perturb a pathway and measure effects at many downstream reporters. Use the measured effects to piece together the pathway structure. This becomes feasible if you assume some simple patterns in the data. NEMs are built on subset relations, like the ones in this example:
Fig.1 shows how pathway structure leaves traces in downstream reporters of perturbation effects. For example, in the left panel the two transcription factors (TF1 and TF2) have targets which will react to their perturbation, but if a common regulator (the kinase) is perturbed the effect will be bigger and include the targets of both TFs as subsets.
Using this idea you can reconstruct the pathway from the observed perturbation effects: for example, if you see two sets nested in a third, you can conclude that you spotted a fork in the pathway (like in the left panel).
Obviously, you don’t want to do all of this by eye and the general idea of inferring pathways from subset patterns can be rigorously implemented in a statistical method:
[A] The pathway hypothesis links the signaling pathway (A-D) to downstream reporters of perturbations (1-10).
[B] In many applications perturbation effects can not be observed at other pathway components (which would make the whole problem much, much simpler).
[C] But they can be observed at the downstream reporters.
[D] The expected effects can then be compared to the observed effects (which are noisy and will contain false positives and negatives) to evaluate how well the hypothesis fits the data.
All of this is implemented and ready to use in the ‘nem‘ R-package.
What’s new with NEMs in 2015?
NEMs elucidate gene regulation in C. elegans
The first recent NEM paper I want to advertise is called Transcription Factor Activity Mapping of a Tissue-Specific In Vivo Gene Regulatory Network by Marian Walhout, Chad Myers and Co in the brand-new journal Cell Systems.
Here, we comprehensively assay TF activity, rather than binding, to construct a network of gene regulatory interactions in the C. elegans intestine.
By manually observing the in vivo tissue-specific knockdown of 921 TFs on a panel of fluorescent transcriptional reporters, we identified a GRN of 411 interactions between 19 promoters and 177 TFs. This GRN shows only a modest overlap with physical interactions, indicating that many regulatory interactions are indirect.
We applied nested effects modeling to uncover the information flow between TFs in the intestine that converges on a small set of physical TF-promoter interactions.
NEMs become Boolean
The second paper called “Analyzing synergistic and non-synergistic interactions in signalling pathways using Boolean Nested Effect Models” is by my former supervisor and NEM co-inventor Rainer Spang.
They make the top-layer pathway (between A-D in Fig 2A) more expressive by including logical functions. In other words, they turn the graph into a Boolean network, but keep the NEM idea that changes can only be observed indirectly through downstream reporters.
Here we describe Boolean Nested Effect Models (B-NEM). This method combines advantages from Boolean Network Models and
Nested Effect Models.
Like Boolean Networks B-NEMs distinguish between the alternative and cooperative activation of a protein, and like normal NEMs, B-NEMs do not need direct observations of protein activity.
Moreover, B-NEMs can use data from assays where several pathway genes are perturbed simultaneously.
Looks like quite a good step forward.