Using treatment response data to find targeted therapies in cBioPortal

The Hyve has implemented treatment support data support in cBioPortal, which has been included in the recently released 3.1.0 version of the platform. This will allow researchers to correlate drug response with genetic alterations, helping them in their search for targeted treatment.

cBioPortal is a web based platform to integrate cancer genomics data from various data types (mutations, copy-number alterations, expression, clinical, etc.) to help oncologists in choosing the best treatment for their patients and researchers in uncovering the mechanisms through which tumors develop. By adding treatment response data to the supported data types, cBioPortal can now also be used to visualize and analyze what happens when an intervention takes place.

What is treatment response data?

There are many ways in which we can intervene in tumor growth and there are different ways to measure response to these interventions. Some examples:

  • in cell-line screening, promising chemical compounds are administered to a large number of cancer cell lines in increasing doses, and the response is recorded as a dose response curve
  • using CRISPR or RNA interference (RNAi), a gene can be knocked out or suppressed to find out if this gene is important in tumor development, and hence could be a target for a new treatment
  • a patient’s tumor cells can be grown in-vitro or in-vivo in a number of models so a number of treatments can be tested concurrently before deciding which treatment to administer to the patient

In the remainder of this blog post we will focus on the example of cell-line screening to illustrate the new functionality.

How is treatment response data stored?

From a dose-response curve, different measures for effectivity of a compound can be derived such as IC50 (concentration of half-maximal inhibition), EC50 (concentration of half-maximal activation) or GI50 (concentration of half-maximal suppression of cell proliferation). These measures are used to simplify the comparison of responses and the identification of correlations. In case a treatment does not evoke a meaningful response, the IC50/EC50/GI50 concentrations may not be reached. In these cases, the notation >x is often used, where x is the maximum concentration that was tried in the experiment. It simply means that we do not know the exact value but we do know that it must be higher than x. Other than that little caveat, the treatment response data simply consists of numbers.

How is treatment response data visualized?

The most commonly used visualization for treatment response data across multiple samples is a so called waterfall plot. A waterfall plot is a bar plot where samples are ordered (usually on the x-axis) by increasing or decreasing value of the outcome variable. By overlaying this plot with mutation data or CNA data you can see whether the responders or the resistant samples are more or less likely to be altered. We have implemented the waterfall plot in cBioPortal as a new plot type.

What can I do with treatment response data using cBioPortal?

Waterfall plots

The most important use case for the treatment response data in cBioPortal is to find correlations between treatment response and properties of the tumor. When there is a correlation between, say, a particular mutation and response to a specific compound, this compound could be a candidate for a precision medicine trial.

Let’s have a look at such a use case in the public cBioPortal. These are the steps:

  1. Go to
  2. Select Cancer Cell Line Encyclopedia (Broad, 2019), and click ‘Query By Gene’. To find the study, it helps to put ‘ccle 2019’ in the search box.
  3. Fill in BRAF in the text box and hit ‘Submit Query’
  4. You will now see the oncoprint with the alterations for BRAF.
  5. Click heatmap and select IC50 in the dropdown menu.
  6. Click the ‘Search for treatments’ textbox and type BRAF.
  7. Hit ‘select all’ and click ‘add treatments to heatmap’.
  8. Note that for a lot of samples there is no treatment data available and that it is hard to see a correlation between BRAF alterations and treatment response. We will use the waterfall plot for that.
  9. Go to the plots tab.
  10. On the vertical axis select Treatment, IC50 and tick Log Scale, and on the horizontal axis select Ordered samples. This will display the waterfall plot.
  11. If you go through the compounds in the dropdown menu, you can see that, for most of them, samples with a stronger response often have a mutation in BRAF.
  12. You can also highlight a specific mutation. Type ‘V600E’ in the ‘Search mutation(s)’ box to see how the BRAF V600E mutation is distributed.

As an example of drug resistance, you could have a look at the gene RB1 and the treatment named ‘Palbociclib’. You will see that RB1 mutants tend to be resistant to this CDK4/6 inhibitor.

It is possible to define a pivot threshold per treatment profile, so per measurement type per study. This pivot threshold will act as the x-axis for the plot. Response values that are below the threshold will be drawn below the axis, values that are above will we drawn above the axis, see image below.

For the the CCLE 2019 study that is loaded in the public cBioPortal, no pivot thresholds have been set, but when you load your own data you can define them. Typically, the threshold is set to a value that if the measurement is below that value a treatment is deemed effective and above it is deemed ineffective.

Generic cBioPortal features


As you have seen in the waterfall plot example, the treatments are selected and added in the oncoprint before they become available in the plots tab. Here the treatments behave in the same way as other heatmap tracks with two differences:

  • in case the required concentration for the measurement (e.g. IC50/EC50) was not reached the value will get a yellow color
  • a pivot threshold can be defined, this will have an influence on the coloring in the heatmap

Currently, the only study with treatment response data is the CCLE 2019 study but it has neither pivot thresholds set nor cases where the required concentration was not reached.

When a pivot threshold is set, this will act as the ‘tipping point’ or value for the neutral color between the two colors in the heatmap, to more easily distinguish between responders and non-responders.

Other plots

In the waterfall plots example we set the horizontal axis to ‘Ordered samples’ to display the waterfall plot, but treatment response data also works with all of the other plot types. For example, you could select Treatment, IC50 and Dabrafenib on the vertical axis and Mutation, Mutation type and BRAF on the horizontal axis. This will produce a box plot that will show you that samples with a missense mutation tend to respond better to Dabrafenib than wildtype samples and samples with other types of mutations. Use the log scale to make this more obvious.

You might now suspect that this drug works for a very specific mutation. This can easily be verified by typing the mutation in the ‘Search mutation(s)’ box. Let’s see what happens when we put V600E there. The samples that show a strong response have a BRAF V600E mutation.

In the plots tab we have also implemented some options to work with cases where the required concentration was not reached. By default, these values are shown but not included in the computations of the plots. They can be included by checking a checkbox. Note that this checkbox only appears when these values are present so you will not see it for the CCLE 2019 dataset that was used in these examples.

Closing remarks

The Hyve provides services to develop, extend and improve features in cBioPortal. Implemented features are released to the community via the cBioPortal repository on GitHub. We hope you like this new feature and it proves useful in your work. Many thanks to AstraZeneca for sponsoring this work. For inquiries on cBioPortal feature development projects or other services around cBioPortal, feel free to contact us.