Skip to content

Latest commit

 

History

History
186 lines (154 loc) · 10.8 KB

tutorial.md

File metadata and controls

186 lines (154 loc) · 10.8 KB
Raw

Integrated network analysis of EGF and IGF stimulation using PHOTON

Make sure to follow the installation instructions. Please create an issue if the instructions are not clear, or if you run into any problems.

The goal of our analysis is to 1) derive signaling functionality scores for the proteins in our network and 2) reconstruct the underlying signaling pathway. In brief, signaling functionality scores describe whether a protein promotes (high positive score) or demotes (high negative score) the phosphorylation of its interactors. We can utilize the proteins with significant scores to extract a signaling pathway from the PPI network that connects the functional proteins.

Data preparation

Phosphorylation data

The standard analysis of PTM data is not covered in this tutorial. Only the most elementary steps are listed here, skipping e.g. data exploration and normalization. The dataset we are going to work with consists of stimulated MCF7 cells measured in duplicates. The samples were measured with SILAC. All reported phosphorylation fold-changes correspond to stimulated/unstimulated condition.

  1. Download the data here.

  2. Load the file phosphoSTY_rudolph.txt (Matrix => Load => Generic matrix upload).

  3. We will eventually map this data to the STRING PPI network. Therefore we need to make sure that the network and the data are defined in terms of the same protein identifiers. Here we choose to covert the UniProt ids in the data to the ensemble protein ids (ENSP) used in the network (Matrix => Processin => Annot. columns => Add annotation). Select mainAnnot.homo_sapiens.txt.gz and annotate with ENSP.

Obtaining a high-confidence mouse protein-protein interaction network from STRING

PPI networks are usually obtained from databases. A popular choice is STRING, which we will be using here.

  1. Download the 9506.protein.links.v10.5.txt.gz file from string-db. It contains all physical interactions between proteins in human.

  2. We start by loading the STRING network table (Matrix => Load => Raw upload). Choose file 9606.protein.links.v10.5.txt.gz and separating columns by 'space'.

  3. The technologies used to generate large-scale interaction networks are not perfect and therefore interaction databases often contain a fraction of false interactions. STRING provides a confidence score, which we can use to filter out low-confidence interactions. First we have to change to column type of the score column (Matrix => Rearrange => Change column types). Convert comined_score from 'Text' to 'Numeric'.

  4. We can now filter the edges and observe the changes in number of edges and score distribution by looking at the histogram of the scores before and after filtering (Matrix => Filter rows => Filter rows based on numerical/main column). Set x >= 900 and create the histogram (Matrix => Visualization => Histogram). Select combined_score and set the y-axis to log-scale.

  5. To make it clearer that the combined_score represents the confidence in each interaction, we will scale its values to the commonly used range between 0 and 1 (Matrix => Basic => Transform). Select combined_score and x / 1000 and then rename the column (Matrix => Rearrange => Rename columns). Change combined_score to Confidence.

  6. In order to improve compatability of the network with UniProt annotations we will change the format of the protein identifiers and remove the species prefix (Matrix => Rearrange => Process text column). Select both columns and set 'Regular expression' to ^9606\.(.*). Make sure not to add any extra symbols or spaces to the regular expression.

  7. The resulting edge table provides the template for our network. We will save this intermediate result to file for use in the next step (Matrix => Export => Generic matrix export). Save as 9606.protein.links.v10.5_high_confidence.txt.

Create the PPI network from the edge table and prepare it for data annotation

  1. If necessary, load the edge table from the previous step (Matrix => Load => Generic matrix upload). Choose file 9606.protein.links.v10.5_high_confidence.txt.

  2. Next to the ‘Matrix' tab Perseus has a 'Network' tab that lists all operations that can be performed on networks. Now we can transform the matrix into the dedicated ‘Network collection’ data structure (Network → From matrix → Basic → From matrix). Choose protein1 and protein2 as 'Source' and 'Target', and check the Network is directed option.

  3. Take your time to browse and understand the different tables that represent a 'Network Collection'. Whenever you select a network collection in the workflow, tables in separate tabs will give you different levels of information on the networks in the collection. Here we have only a single network, which is represented by a single line in the Graphs tab. Next to the Graphs tab you can select the tab of the network, named accordingly. Each network is itself represented by two tables. The 'Nodes' table provides details on all the molecules that are part of the interaction network and will always contain a Node column that identifies each entry. The 'Edges' table describes all interactions between the nodes and always contains at least a Source and Target column. Keep coming back to these tables in order to see the effect of any of the next processing steps.

  4. We can explore the topology of the network by calculating node degrees (number of neighbors of any given node, Network => Topology => Node degrees).

  5. The node degrees were added to the 'Node' table of the network. Make sure you can find them. It is easiest to plot them using any of the matrix visualizations in Perseus. We can easily convert network tables to matrices (Network => To matrix => Basic => To matrix). Select 'Nodes'.

  6. When we plot the histogram of node distributions we can see that there are a number nodes with very large degrees (Matrix => Analysis => Visualization => Histogram). Select Degree and change the y-axis to log-scale.

  7. Finding out the identity of the proteins by browsing ensemble protein identifiers is tedious. Instead, we would like to map them to gene names. We can't do so directy, but first have to map ENSP to UniProt identifiers (Matrix => Processing => Annot. columns => To base identifiers). Select mainAnnot.homo_sapiens.txt.gz and 'Identifier type' as ENSP. Now we can add annotations as usual (Matrix => Processing => Annot. columns => Add annotation). Select mainAnnot.homo_sapiens.txt.gz, set 'UniProt column' to UniProt and add the Gene name and GOBP name annotation.

  8. Sorting by degree will show you the highest connected proteins in the network. Due to their high connectivity, These proteins can have a dominating effect on network analyses. Therefore, unless these proteins are of interest, removing them from the network collection can be helpful (Network => Processing => Filter nodes => Filter nodes by numerical column). Enter x < 1000.

  9. [Optional] This preprocessed intermediate result can be saved to the hard drive as a folder (Network → Export → Generic network export). Create a folder named 9606.protein.links.v10.5_preprocessed at your desired location and select it as save destination. Use the file explorer to navigate to the save location and open the network folder. You will see a series of .txt files that represent the network tables you were browsing in Perseus. You can inspect and edit these tables with any decent text editor (Notepad might not be able to handle the files due to size). Due to the simple tab-separated format, it is easy to import the network into other tools, such as Cytoscape.

Perform PHOTON analysis and visualize the results

  1. We are finally ready to perform the analysis. Make sure that the data preparation was successful and load intermediate result from file if necessary.

  2. To annotate the network with the experimental data, first select the network collection in the workflow, then, while holding the CTRL key, select the data table. With both workflow items selected, perform the annotation (Network → Merge with matrix → Annotate → Annotate nodes). Match Node with ENSP and copy all main columns. In order to not collapse the peptide-level information of the data to the protein-level network, we set 'Combine copied main values' to keep separate.

  3. Make sure that the node table of the network contains multi-numeric columns with the data.

  4. Perform PHOTON analysis to obtain 'Signaling functionality scores' for all proteins in the network. (Network => Processing => Modifications => PHOTON). Select the Reconstruct signaling networks using ANAT option and enter ENSP00000265171 (ENSP id of EGF stimulus) as signaling source.

  5. For convenience the result is listed in different data nodes. The big network collection contains all results of the analysis. The second network collection contains reconstructed signaling networks. The matrix contains all signaling functionality scores which can be analyzed further, e.g. by clustering (Matrix => Processing => Annot. columns => To base identifiers). Select mainAnnot.homo_sapiens.txt.gz and 'Identifier type' as ENSP (Matrix => Processing => Annot. columns => Add annotation). Select mainAnnot.homo_sapiens.txt.gz, set 'UniProt column' to UniProt and add the Gene name and GOBP name annotation (Matrix => Analysis => Hierarchical clustering). The time points will cluster together. You can flip branches of the cluster dendrogram by CTRL+click on the branching points. Can you identify some interesting cluster.

  6. You can create a basic visualization of the reconstructed signaling networks directly inside Perseus. First add gene name annotations to the network (Network → Processing → Annot. columns → To base identifiers). Select mainAnnot.homo_sapiens.txt.gz and ‘Identifier type’ as ENSP. Add name and GO annotations (Network → Processing → Annot. columns → Add annotation). Select mainAnnot.homo_sapiens.txt.gz, set ‘UniProt column’ to UniProt and add the Gene name annotation. Visualize the network as a node-link diagram (Network → Analysis → Visualization → Kinase-substrate network). Select Gene name as ‘Node column’ and From network name as ‘Data column’.

This covers the basic analysis. The kinase functionality scores can be analyzed in a similar way to protein quantification, i.e. via clustering, statistical testing etc.