Florian Markowetz, Dennis Kostka, Olga G. Troyanskaya, and Rainer Spang. Nested effects models for high-dimensional phenotyping screens. In Proceedings ...
Nested Effects Models for High Dimensional Phenotyping Screens Achim Tresch Practical DNA Microarray Analysis, Dortmund, September 2007
This exercise will give an introduction to the use of nested effects models (NEMs). There are two packages which implement different variants of nested effects models: nem which was introduced in Markowetz et al. (2005, 2007) and Nessy from Tresch and Markowetz (2007). The former can already be obtained from Bioconductor, the latter will be avaliable soon. It can be downloaded from http://www.staff.unimainz.de/tresch/nessy.html 1.) A toy example. Use the package Nessy. > library(Nessy) Let us generate and display a randomly generated Nested Effects Model first. > nem1 = createnem(actions = 4, edges = 4, dag = T, observables = 50) > plotnem(nem1, main = "True Model") True Model
2
3
1
4
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
4 3 2
4 8 12 16 20 24 28 32 36 40 44 48 1 5 9 13 17 21 25 29 33 37 41 45 49 2 6 10 14 18 22 26 30 34 38 42 46 50 3 7 11 15 19 23 27 31 35 39 43 47
1
Figure 1. The bold dark grey arrows show the actions graph, the light grey arrows display the effects graph. The heatmap below is a color/grey scale coding of the ratio matrix
The posterior Score of this NEM can be obtained by > posterior(nem1) Once you have created a NEM object, you can search for a model which fits the underlying data (the ratio matrix) best > nemexhaustive = fit.nem(nem1, method = "exhaustive") > posterior(nemexhaustive) We are mainly interested in the actions graph of a NEM. It can be displayed separately in a more convenient layout
> plotactions(nemexhaustive, main = "Exhaustive Search")
Exhaustive Search
2 1
3 4
Figure 2. This graph is a reconstruction of the original actions graph in nem1 from the noisy ratio matrix in nem1. Therefore, the actions graph of nemexhaustive and of nem1 should ideally be identical.
2.) Analysis of a real data set. We apply NEMs to a microarray dataset of an RNAi screen in Dosophila Boutros et al. (2002). The Drosophila dataset is contained in the nem package. > library(nem) > data(BoutrosRNAi2002) For simplicity, we restrict our attention to four genes that have been knocked down in this experiment: tak,key,rel,mkk4/hep. The task is to reconstruct the structure of the signalling cascade of these actions. It is noteworthy that the signal transmission happens on the protein level, and not on the transcriptional level! The data is provided in datamatrix as a genes(=observables) x actions matrix of log intensities. > actions datamatrix ratio + + > > + >
selectedgenes = which(apply(ratio, 1, function(x) { sum(x > 0) > 1 })) nem2 = ratio2nem(ratio[selectedgenes, ]) nem2fit = fit.nem(nem2, method = "exhaustive", featureselection = T, minobs = 30, trials = 5, verbose = T) plotnem(nem2fit, main = "Estimation by NEM")
7.1
Estimation by NEM
3.9
tak−
Graph from the Literature
0.6
1.9
key−
mkk4hep−
tak−
−0.21 −2.9
c(−1, 1)
rel−
●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●
mkk4hep−
rel−
−4.5
tak−
−6.1
rel−
mkk4hep−
key− −8.5
key−
Figure 3. Comparison of the NEM estimation procedure and the literature graph.
References Michael Boutros, Herv´e Agaisse, and Norbert Perrimon. Sequential activation of signaling pathways during innate immune responses in Drosophila. Dev Cell, 3(5):711–22, Nov 2002. Florian Markowetz, Jacques Bloch, and Rainer Spang. Non-transcriptional pathway features reconstructed from secondary effects of RNA interference. Bioinformatics, 21(21):4026–4032, 2005. Florian Markowetz, Dennis Kostka, Olga G. Troyanskaya, and Rainer Spang. Nested effects models for high-dimensional phenotyping screens. In Proceedings of ISMB/ECCB, 2007. Achim Tresch and Florian Markowetz. Structure learning in nested effects models. submitted to Biostatistics, 2007.