First, we run the
prePHONEMeS_rawData2PHONEMeS.R script from which we naormalize our raw data (MSdata.RData), perform some statistical analysis of the normalized data by applying a linear model (see Limma package) thus obtaining the significance of the fold changes between the perturbed and control measurements. Then a Gaussian Mixture Model is applied (see mclust R package) in order to identify those peptides which are best fitted with a certain number of components/clusters. In addition we remove those peptides where the control is NA since we cannot assign a control cluster on this case. Only those measurements which are best described with two comoponents are taken into consideration. A probability is given to each of the peptides on each of the condition based on the resulting GMM parameters which tells how likely is this measurememnt to belong to the control or the perturbed state. The output of this script will be a data matrix
GMM.res where we identify categories of data points after linear model estimation of fold changes vs control and Gaussian mixture modelling by associateing to the fold changes for drug vs control an adjusted p-value which is estimated with the linear model.
preOptim.R which takes as an input the data matrix output from
prePHONEMeS_rawData2PHONEMeS.Rand a frame of data containing all possible K/P-Substrate interactions that we can retrieve from large data-bases such as Omnipath (see original publication of Omnipath here). After defining our target kinases(MTOR on this case), the script will choose the drug treatments matching to the drug targets and match to what is present in the background network. This will allow the building of a background prior knowledge network containing all possible interactions connecting our drug targets with the measurements we get. The
preOptim.R will then create the objects needed to run PHONEMeS on a cluster and the optimization parameters such as:
ndenotes the index of the independent optimization run.
Before running the analysis on the cluster, we have to make sure that the PHONEMeS package is already installed on the cluster. After we make sure that PHONEMeS has been successfully installed, we import on the cluster the outputs from the previous step
data4cluster_n.RData and the other following scripts (important note: in the cluster we create different directories for each independent run through
processGx.Rwhich combines all the results (model and scores) for each independent optimization runs.
runScriptGx.shfor running the parallel processes
scriptGxopt_50models.Rwhich optimizes all the models sampled.
import.Rfor locally importing the outputs from the analysis on the cluster. After running the bash script
./runScriptGx.sh n, on each of the folders we created we will have our outputs which we need to import locally.
After every result has been imported successfully, we process our results with
postOptim.R which will assign attributes to each of the nodes and edges of each of the independent runs and save everything on one single folder (
comb_optim.R we will then combine all the
objects_pn.RData in a sngle final resulting network.
As for the visualization of the resulting network, we can use Cytoscape. For more details check the description in the Usage section.