Performing Hierarchical Cluster Analysis

Interpreting Hierarchical Cluster Analysis

Performing Partition Cluster Analysis

Interpreting Partition Cluster Analysis

Entering Protein Names into EDA

Querying Protein Databases for more Information

  

-Preparing for EDA Analysis 

• EDA Analysis is performed by combining several BVA workspaces. In order to ensure a smooth import of data from EDA to BVA, the following conventions should be kept in BVA: 

• Use the experimental design block in BVA to put each spot map into its proper experimental group (Standard, Control, Treatment_1, etc.) Maintain consistency of these groups for each BVA workspace to be analyzed by EDA 

• Select one "master" gel from the entire experiment and use it to match the gels in each of the BVA workspaces. For the BVA workspace that already contains this gel, import it a second time and match all other gels to it, including itself. For the experimental design, the spot maps for this gel should be put in the "Unassigned" group in all cases. 

• The master gel must remain the same in all BVA workspaces!! This means that if two spots are merged on the master gel in one BVA workspace, they must also be merged on the master gel in all other BVA workspaces, whether they match the other gels or not.

 

-Creating an EDA Workspace 

• In the "Step 1 - Workspace" section, select Create Workspace. 

• Select the proper project in the first column of the menu that pops up. 

• Select the BVA workspaces in the second column to be analyzed in EDA. 

• Click the Add button, then click Create.

 

-Creating the Base Set 

• If all gels in the entire experiment (across all BVA workspaces) used one common internal standard, select Automatic in the "Step 3 - Base Set" section at the bottom left and then skip to Filtering the Data. 

• If all of the gels did not share one common internal standard, select Manual in the "Step 3 - Base Set" section at the bottom left to normalize the spot maps. 

• Click the normalization tab at the top of the menu. 

• Select any experimental group common to all the workspaces (e.g., Control) in the first pull-down menu. Select the workspace where the master gel is also assigned to an experimental group as the reference workspace for the second pull-down menu. Click Apply Normalization. It will take at least a few minutes for the normalization to complete. Close this window when completed.

• In the Experimental Design section in the right column, select one of the experimental groups, click Edit Group, and assign it to a color by clicking on the color block. Do this for each experimental group,assigning each group a different color. 

 

-Filtering the Base Set (optional but recommended) 

• Select Calculations in the workflow diagram at the top of the screen. 

• Select Filter Set in the left column. 

• In the Protein Filter section, select the dropdown criteria "% of spot maps where protein is present", select ">=", and input a value for the cutoff. For example, if 75 (recommended) is used, EDA will exclude any proteins found in less than 75% of the spot maps. 

• Click Add. 

• In the Spot Map Filter section, select the dropdown criteria "Remove unassigned spot maps". Click Add. 

• Click Apply Filter. 

• Click Create Set at the bottom of the window, and give the new set an identifying name, such as "Filtered Set". 

Note: Filtering the base set is not mandatory but will produce more robust and consistent analysis. 

-Switching between Sets 

• To switch between sets in any menu, click the dropdown menu under "Select Set:" in the left column, choose the desired set, and click Select Set. 

 

-Calculating Differential Expression 

• Click the Calculations button in the workflow chart at the top of the page and switch to the Filtered Set. 

• Select the Differential Expression Analysis calculation. 

• Create settings for the calculations as follows: 

o       If the experiments compare the same patient/animal under different conditions, select "Paired tests"; otherwise, select "Independent Tests". 

o       If you are interested in comparing two experimental groups (e.g. Control and Treated), check the Average ratio and Student's t-test boxes. Select the desired groups (use Ctrl-click to combine multiple groups), name the calculation, and click Add to List.

  o       If you are interested in looking for differential expression among multiple experimental groups, follow the directions above but also check the One-way ANOVA box and check the "Calculate multiple comparison tests" box. 

• Select Calculate.

 Note: Only one differential expression calculation is accessible at any time for a given set. Creating a new calculation will delete the old calculation.

-Viewing the Differential Expression Results

• Click Results in the workflow diagram and click the Differential Expression Analysis tab. The lower section of this page, shown below, is a table with tabs for the proteins and the spot maps. The Protein tab contains the average ratio, t-test score, and ANOVA score for each protein. To view the multiple comparison tests (if any), select the protein in the table. The multiple comparison test results will be shown in the upper section of this page, along with a plot of the log standard abundances for each spot map, split up by experimental group.

• The protein table can be sorted by clicking the header of the column by which you wish to sort (e.g. click "T-test" in the table to sort the proteins from highest to lowest t-test score; click it again to sort from lowest to highest).

• The settings for both the table and the plot can be altered by clicking the Settings Dialog Box as shown in the picture shown below.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

-Filtering the Data by Differential Expression

• You should now filter the proteins to include only those that show differential expression for further analysis. The proteins can be filtered by either t-test and/or ANOVA. They can also be filtered by average ratio, but this should be used only in addition to ANOVA and/or t-test, and not in place of them.

• While still in the Results/Differential Expression Analysis tab, select Filter Set in the right column. Under protein filter, select a filter criteria from the dropdown menu and set the desired threshold for statistically significant differential expression (e.g., "Student's T-test", "<", 0.05". Click Add.

• Add as many filters as desired in the same manner. Use the "AND all" radio button to only include proteins which satisfy all filter conditions; use the "OR all" radio button to include proteins which satisfy any one of the filter conditions.

• To remove a filter, select the condition and click Remove.

• Click Apply Filter.

• Click Create Set and give the set an identifying name and color, such as "Differential Expression".

 

-Performing Principal Components Analysis (PCA)

• PCA uses plots to identify protein outliers, if any, and to confirm that protein expression is consistent throughout multiple samples from the same experimental group.

• Click on Calculations in the workflow diagram.

• Switch to the set containing the differential expressed proteins.

• Click on Principal Components Analysis under "Select Calculation" in the left column.

• Select the upper left radio button, which plots proteins against spot maps. Give this calculation an identifying name, such as "Proteins - Spot Maps" and click Add to List.

• Select the lower left radio button, which plots spot maps against proteins. Give this calculation an identifying name, such as "Spot Maps - Proteins" and click Add to List.

• Click Calculate in the right column.

 

-Interpreting the Principal Components Analysis: Proteins - Spot Maps

• The Proteins - Spot Maps calculation will perform statistical checks to identify if there are any protein outliers, i.e., spots that have been mismatched or are not true proteins).

• For most purposes, only the plot on the left side is important, and is thus the only plot shown below.

• Each dot represents a protein. The circle on the plot represents a 95% confidence interval within which the expression of a true protein is expected to lie based on the general expression patterns in the spot maps. Any protein that falls outside this interval, like the one indicated by the red arrow above, should be checked in BVA to make sure it is properly matched and a true protein. There is not necessarily anything wrong with this protein, however; good proteins that have especially strong differential expression often end up outside the 95% area.

• To look at a protein in BVA, click on the dot, click on a spot map in the table, and go to Tools à Open Source in the menu bar.

 

• If any proteins are actually artifacts or are mismatched, exclude them from the set by Ctrl-clicking to select them in the plot and then clicking Create Set in the left column. Select the radio button for Removing Selection in the Proteins section, and select the radio button for Including All in the Spot Maps section. Give the new set an identifying name (such as Differential Expression - Filtered) and click Create.

 

-Interpreting the Principal Components Analysis: Spot Maps - Proteins

• The Spot Maps - Proteins calculation will plot the spot maps on a graph based on the expression patterns of the selected proteins. It is used to make sure that multiple trials of the experiment are producing consistent results.

• Once again, the important plot is the one on the left hand side of the results, so this is the only plot discussed here and shown below.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

• Each dot on this plot represents a spot map, based on the expression of the proteins in the set used for the calculation (this should be the Differentially Expressed - Filtered set). Each dot will be colored with the color of the experimental group it belongs to, assuming that the groups were assigned unique colors in the experimental design (see "Creating the Base Set").

• If the experiment is producing consistent results, spot maps from the same experimental group should be located in the same general area, i.e., contained mostly in one quadrant or half of the plot.

• If a spot map is located outside the 95% ellipse, or is located far away from the other spot maps in its experimental group (an example is indicated by the red arrow in the picture above), then you may want to exclude this spot map from further analysis. It is recommended to look at the gel in your imaging software. If there are widespread gel abnormalities or streaking, the gel should be excluded in the same manner as the proteins were in the previous section. Otherwise, it is acceptable to keep the gel in the analysis.

• This calculation can also be used to make some rudimentary conclusions about the experimental groups. If the locations of the spot maps in two or more experimental groups overlap a great deal (as with the green and orange spot maps in the image above), then the samples in these groups show few, if any, differences in protein expression, and the groups are likely very similar. On the other hand, if the spot maps in two or more experimental groups are found in mostly distinct areas of the plot (as with the blue and red spot maps in the image above), these samples show many consistent differences in protein expression, and the groups are likely fundamentally groups. However, these conclusions can and should be refined with further analysis.

 

-Performing Hierarchical Cluster Analysis

• Hierarchical Cluster Analysis will group proteins together in a hierarchical tree by analyzing the expression patterns of the proteins in the different spot maps. Proteins showing similar expression patterns will be located near each other in the tree, while those with different expression patterns will be farther away. This can also be done with spot maps. The result of this analysis is a heat map that will be discussed in the following section, "Interpreting Hierarchical Cluster Analysis".

• Click Calculations in the workflow diagram at the top of the page. Make sure that the filtered set of differentially expressed proteins is selected as the current set.

• Click Pattern Analysis in the left column.

• Select the Hierarchical Clustering Algorithm.

• Select the upper left radio button to cluster the proteins based on their expression in the spot maps.

• Give the calculation an identifying name, such as "Proteins", and click Add to List.

• Select the lower left radio button to cluster the spot maps based on the expression of the proteins in them.

• Give the calculation an identifying name, such as "Spot Maps", and click Add to List.

• Click Calculate in bottom of the right column.

 

-Interpreting the Hierarchical Cluster Analysis

• Click Results in the workflow diagram at the top of the screen. Then click the Pattern Analysis button and click the Hierarchical Cluster Analysis tab.

 

• The image above shows a screenshot of the Hierarchical Cluster Analysis results. In the heat map, each horizontal row represents one protein (with the protein number at the end of the row). Each vertical column represents one spot map (with the details underneath the column). As a result, each rectangular box, such as the one highlighted in pink in the image, represents the expression of that row's protein in that column's spot map as compared to the internal standard. A green color means that the expression decreased compared to the standard, and a red color means that the expression increased compared to the standard - the brighter the color, the stronger the change, as seen by the scale to the lower left of the heat map.

• If the cursor is held over a box, the numerical change in expression will appear under the heat map scale.

• The trees to the left of the heat map and above the heat map show the relationships between the proteins and between the spot maps, respectively. Proteins that split off into different parts of the tree near the root of the tree are less related than those that split off near the heat map. For instance, in the image above, the tree at the left splits the proteins near the root into two main groups. Looking at the tree, the protein indicated by the pink box is the last protein in the first group, while the proteins below it all belong to the second group.

• Spot maps can be looked at in the same way. Generally, one would expect spot maps from the same group to be clustered closely together, as is the case with almost all of the red "control" spot maps in the image above. As with the PCA, this is a useful tool to ensure that the experiment is producing consistent results, and it can also be used to identify possible cases of individual variation.

• The buttons to the upper right of the heat map allow the user to change some settings. On the top row, the leftmost button will show the view shown above and the middle button will shown only the heat map (and make it bigger). The rightmost button allows the user to change the scale of the heat map, which is visually useful if most changes are only on the order of 50%. The bottom row of buttons allow the user to zoom in or out on the heat map.

 Note: The image above for the hierarchical cluster analysis comes from a different experiment than the image for the PCA. Otherwise, each spot map shown on the PCA image will correspond to a particular spot map in the hierarchical cluster analysis - these can be matched using the Spot Map table in the bottom half of the screen if desired. 

-Performing Partition Cluster Analysis

• Partition Cluster Analysis groups proteins into clusters that show similar expression patterns across the different experimental conditions. Spot maps or experimental groups can also be clustered but this is not useful for most experiments.

• There are three methods for partition cluster analysis. The first, Kmeans clustering, generates a certain number of disjoint clusters based on the spread of the data. The second, Gene Shaving, generates a large number of small clusters, more than one of which can contain the same protein. The third, Self-Organizing Maps, tends to generate a small number of large clusters, and is not preferred.

          K-means

• Select the Kmeans Algorithm.

• Select the upper left radio button to cluster proteins according to expression in spot maps.

• Give the calculation an identifying name, such as "KMeans", and click Add to List.

• Click Calculate at the bottom of the right column.

 

      Gene Shaving

• Select the Gene Shaving Algorithm.

• Select the upper left radio button to cluster proteins according to expression in spot maps. 

• Give the calculation an identifying name, such as "Gene Shaving", and click Add To List.

• Click Calculate at the bottom of the right column.

 

-Interpreting Partition Cluster Analysis

• Click Results in the workflow diagram at the top of the screen. Then click the Pattern Analysis button and click the Partition Cluster Analysis tab. Use the pulldown menu to select the desired calculation (Gene Shaving, KMeans, etc).

• An example screen of the results from partition cluster analysis is shown below. Summaries of each cluster (expression pattern, number of proteins, and "q" score) are displayed in small boxes in the main window. A detailed chart of the expression of each protein in the cluster is also shown in the main window. A table showing each protein in the selected cluster is located below the main window. 

• The "q" score, shown above each of the clustery summaries, is a measure of the similarity between the expression of all the proteins in the cluster. The maximum q-score is 100, so the closer the q-score is to 100, the stronger the pattern between all of those proteins. The q-score has no meaning as a comparative value between different clusters, though; it is merely an expression of the precision of an individual cluster.

• Proteins that are in the same cluster can be significant in several ways. First, if proteins that are next to each other are clustered together, this indicates that these are the same protein, and strengthens the belief that this protein is differentially expressed. Second, if proteins show similar expression patterns, it is possible that they either serve similar functions, are affected by similar substances or events, or belong to the same biochemical pathway. All of these possibilities should be researched and analyzed further. 

-Entering Protein Names into EDA

• Proteins that are candidates for further analysis, both in EDA and in future experiments, should be picked from a gel and identified via mass spectrometry. Once these proteins are known, their names and UniProt/NCBI accession numbers should be entered into EDA.

• To enter the name of a protein, click on it in a table and then click Select in the box to the left of the table. A new window will open. 

• Click Add.

• Type the name of the protein in the text box underneath name.

• Type the accession ID in the text box underneath the appropriate database. If the format of the ID is not valid, the text box will be outlined in red and you will be unable to enter that ID.

• Click Add.

• Click Close.

 

-Querying Protein Databases for More Information

• EDA has several tools that can be used to quickly acquire information from protein databases and other sources about the proteins you have identified.

• Click Interpretation in the workflow diagram at the top of the screen.

• Be sure that the final set of proteins is selected.

• Click Create Query and select the desired query.

o       Select Gene Ontology to view the molecular functions, biological processes, and cellular components associated with the identified proteins based on the genetic motifs they contain.

o       Select Pathways to find any commonly recognized pathways in which any of the identified proteins take part.

o       Select UniProt Features to obtain a summary table with general information about taken from UniProt about each of the identified proteins. This information includes function, pathways, family, and cellular location.

o       Select PubMed to retrieve information and articles about the identified proteins from PubMed. This is only available with a connection to discoveryHub.