8. Working with Kaplan Meier¶
Investigate the prognostic value of a gene or a group of genes
8.1. Scope¶
- Use R2 to generate a Kaplan graph by “annotated parameter”. Use tracks or combine two tracks to assign the group separation of a specific dataset.
- R2 supports any type of survival data, such as overall survival and relapse free survival.
- Use the Kaplan Scan for a group of genes.
8.2. Step 1: Selecting the Kaplan Meier module¶
Logon the R2 homepage and select the option R2: Survival (Kaplan-Meier/Cox) either in the left menu panel on the main screen or in field 3 at the type of analysis pull down menu. Using the Kaplan Meier module via the left menu directly shows from which datasets survival data is available.For this example, make sure that Data type is set “Tumor Neuroblastoma public “ Versteeg “ 88”. From the dropdown of the setting Separate by choose the option ‘a categorical track’. Click Next.
In the adjustable settings menu choose for Type of survival the value “overall-c1103”, select “track” at Separate by and select “inss (5 cat)” at the Track pull-down menu . Click “Next”. Note that stage st4s en st1 survival curves are overlapping which is in agreement with the clinical outcome of the INSS stages.
A handy feature of the R2 Kaplan module is the option to combine two tracks to generate subgroups for the Kaplan Meier analysis. Use the Kaplan-start button (top-left), click ‘Next’. Choose overall-c1103 again and select at Separate by the option ‘combination of two tracks’. For example, choose for the first track ‘agegroup (2 cat)’ and for the second track ‘mycn_amp (2 cat)’. And click Next.
The combined track agegroup ( >18 year) and no mycn application results in intermediate survival probability. Note that there are 3 groups instead of “expected” 4 since there are no patients <= 18 year and a mycn amplification, in this cohort.
Did you know that you can apply a filter to KaplanScan analyze a subgroup of patients for survival? In addition you can also adapt the graphical representation*
You can either use the annotation grid (left) or you can use the pull down (right) to select a subset of your samples by one or more tracks. If you click on the subscription of the picture above, the picture will open in large.
For the use of the grid, use the clog icon in the ‘Advanced settings box’ and hover your mouse over the side of the column of choice to show the filering options for that track. You can use the filtering options of multiple tracks to make a specific subset. Here we used gender to only select male samples (i.e. deselect female) and death_cause to select tumor and nd samples. A successful subset selection will be shown as a small message indicating the number of samples in the subset, the used track name and the number of groups/categories between brackets
If you instead want to make the subset selection with the pull-down of Subset track, click on the track by which you want to define the subset (in the example: gender). In the popup you need to check the box of your preferred subset(s) (here: male).If you want to further narrow down your selection with a different track, click on the same pull-down menu. Select the next track (here: death_cause) that you are interested in and in the popdown, check the preferred subset(s) from that track (here: tumor and nd).
Don’t forget to click on Redraw Graph after you made your final selection to redraw the Kaplan Meijer curve.
Nb. If you use the ‘back’ button in your webbroswer, then this selection will be lost and needs to be defined again.
8.3. Step 2: Kaplan Meier by gene expression; the Kaplan Scan¶
An often used feature of R2 is the Kaplan Scan (KaplanScan), where an optimum survival cut-off is established based on statistical testing instead of, for example, just taking the average or median. The Kaplan scanner separates the samples of a dataset into two groups based on the gene expression of one gene. In the order of expression, it will use every increasing expression value as a cutoff to create 2 groups and test the p-value in a logrank test. The highest value is then reported, accompanied by a Kaplan Meier picture. So in short, it will find the most significant expression cutoff for survival analysis. The best possible Kaplan Meier curve is based on the logrank test. However, R2 does also allow you to use median, average and more as a cutoff in assessing whether a gene of interest has the potential to separate patient survival. Of course, such analysis is only possible for datasets where survival data is present.
Select from the main screen either the left menu the Kaplan Meier option or in field 3 of the main menu. Click Next. Make sure that “Tumor Neuroblastoma public Versteeg 88” is selected with Data set ‘Expression data (H. sapiens) and select in the Separate by menu: “single gene (Kaplan-Meier)” and click next.
In the next screen fill in ‘mycn’ in the Search by Gene field and use the first reporter in the list of the dropdown that popspup. Select “overall-c1103” at type of survival and click “Next”.
The Kaplan scan generates a Kaplan Meier Plot based on the most optimal mRNA cut-off expression level to discriminate between a good and bad prognosis cohort.
The determined separation in groups can be stored in a track and used in other analyes, click the “store as track” button
Go back to the plot screen which should still be open. To illustrate that with the Kaplan Scan more significant biological subgroups can be found, adjust the cut-off mode to “median” in the settings menu and click Redraw Graph.
Figure 6: Kaplan plot with multiple cutoffs: A) Scan B) First Quartile C) Median D) Average
It is obvious that with the Kaplan Meier “scan modus” the sample grouping is much more significant compared to the median cut-off modus. Try to find out whether this is also the case with other cut-off modus
The animated gif in figure 5 shows that in the gear box the scan plot can be depicted which represents a graphical representation of the p-value plotted against the mRNA expression level values. In some cases it could be useful to change the p-value cut-off level and for this reason this graphical p-value plot (which is clickable) could be of help. Alternatively, you could use the “cutoff” field to regenerate a Kaplan curve with that separation.
8.4. Step 3: Kaplan scan for a group of genes¶
Instead of using the Kaplan Scan for a single gene you can also analyse a group of genes at the same time. Click on Kaplan start and choose Separate by ‘multiple genes’ and click “Next”.
In this example search for a Gene set by clicking on “select gene set”. In the popup grid, type ‘apoptosis’ in the text field on top and click on the search button. In the adapted grid, open the KEGG pathways by a click on the arrow. Scroll and go on untill you can check the KEGG ‘Apoptosis’ pathway. Click Confirm Selection to go back to the menu, where the ‘GS: Apoptosis (85)’ geneset is now shown to be selected.
Set the Type of survival at ‘overall-c1103’. In the statistics panel there are several filtering options possible, leave these options unchanged.
In the graphics Settings section select “yes” at Draw heatmap and click Next.
In the next screen R2 has generated a list of the genes within the apoptosis pathway which have significant prognostic value. A heatmap for this list of genes is generated as well.
In Figure 8B, the result is shown when clicking on a gene name in the hugo column. It is shown in a new screen or tab with the corresponding Kaplan plot.
In case the Draw heatmap is set to Yes heatmap directly plotted next to list of found genes with significant cut-offs. Clicking a spot in the heatmap will show the gene expression level directly for all samples via a new one-gene-view screen.
Figure 9: Heatmap of the significant prognostic list of genes.
- To generate a binary heatmap based on the GOOD versus BAD prognoses, again go to the main page, click on Kaplan-Meier from the left menu, select ‘by multiple genes’ in the Separate by option, click next. Set Type of Survival to overall-c1103 again. This time don’t select anay gene set. In the Graphic Settings set Draw a heatmap again to ‘yes’, choose in Heatmap data dropdown ‘good_bad (binary)’ and click Next.
Figure 10: Select binary heatmap.
Now the heatmap shows a clustering based on the GOOD vs BAD prognoses. (Figure 11)
8.5. Step 4: Kaplan scan on your own cohort¶
It may happen that you would like to use the KaplanScan method on a dataset that is not available in R2. Especially for this reason we have made a user defined version within R2. Here you can paste your cohort into R2 from e.g. a textfile and run the procedure. To initiate such a user defined Kaplan Scan, select the Kaplan start option again from the main page left menu, and click on the Kaplan-Meier analysis using custom data button underneath the panel.
For the remaining steps to work as intended, we need to take into account a couple of things. You should prepare your data in the following four tab- or semicolon(;) separated columns.
- Column 1 contains a sample identifiers (without spaces)
- Column 2 contains the survival time in days (R2 will convert these)
- Column 3 contains the censoring information (event) and can be yes/no or 1/0
- Column 4 contains the expression value of the gene of interest for the kaplanscan
You could easily prepare this information in Microsoft Excel and paste the selected columns into the large white paste box. Do take care that we use “.” for decimal signs.
After you pasted the dataset information, you make the selection for the cutoff option and subsequently press next. R2 will now calculate the kaplan method that you selected and display the result in an interactive image.
Once the image has been created, you are able to adapt various parameters to optimize appearance of your result.
Did you know that the survival data used in your scan produces a unique signature?
R2 will indicate within the image a checksum (MD5 sum) of all the survival information, which can be used to identify whether the same cohort information has been used in different scans that you may perform (this code should remain identical).
8.6. Step 4: Cox Regression analysis and hazard ratio¶
The Cox regression analysis is a statistical method commonly used in biomedical research to analyze the relationship between gene expression and survival outcomes. It is particularly useful for studying the impact of gene expression levels on patient survival times. In this analysis, gene expression data is combined with survival data to assess whether specific genes or gene signatures are associated with increased or decreased survival rates. In r2 you can identify genes or multiple genes (genesets which may act as potential prognostic markers). In general Cox regression analysis in gene expression provides valuable insights into the molecular mechanisms underlying survival outcomes in various diseases, including cancer.
Figure 14: Cox regression and hazard ratios
The hazard ratio (HR) is a fundamental concept in Cox regression analysis. It represents the relative risk or likelihood of an event occurring in one group compared to another. The hazard ratio quantifies the effect of gene expression levels on the hazard or risk of a specific outcome, typically survival.
Figure 15: Cox regression and hazard ratios steps
In R2 you can scan for significant hazard ratios for a single gene or multiple genes for a single dataset, follow the steps as indicated in Figure 15 and select the drug target geneset and click next. A Hazard ratio quantifies the effect of gene expression levels on the hazard of risk for a typical outcome. When the Hazard ratio is greater than 1, higher expression levels suggest a poor prognosis indicated in blue. Having the opposite, a hazard ratio less than 1 indicates a decreased risk or hazard, suggesting a better prognosis indicated with red. A hazard ratio of 1 implies that there is no difference in risk between the groups being compared.
Figure 15: Cox regression and hazard ratios list
The hazard ratio outcome can easily be inspected by clicking on the KAPLAN” link in the table. As clearly depicted in figure 16 the CSKN2A1 gene with a hazard ratio > 1 in blue show a poor prognosis for a high expression level cut-off while the NTKR1 gene shows a poor prognosis for a low expression level.
Figure 16: Cox regression and hazard ratios list
After checking the hazard ratio for multiple genes in one dataset you can also check for hazard ratios across multiple datasets. Select via de main menu survival > cox regresssion in multiple datasets. (Figure 17). Clicking “select datasets” will open de grid where you select the datasets of interest, keep in mind that these dataset are already preselected for those containing survival data.
Figure 17: Cox regression for multiple datasets
In Figure 18, the selected datasets show a significant low hazard ratios with a poor survival for the NOTCH2 gene in the low gene expressed group in contrast to the hazard rations where the group with the high MYCN expression shows a poor prognosis. (Figure 19.)
Figure 18: Hazard ratios for the NOTCH2 gene
8.7. Final remarks / future directions¶
Everything described in this chapter can be performed in the R2: genomics analysis and visualization platform (http://r2platform.com or http://r2.amc.nl)
We hope that this tutorial has been helpful, the R2 support team.