1. Define how many groups to compare
  2. Assign tracks to groups
  3. Run hypothesis test
  4. Configure P-value track
  5. Get regions with significant P values
  6. Genome-wide hypothesis test

The hypothesis test

The hypothesis test compares groups of heatmap tracks and finds regions of difference with statistical significance. P-values from the test are rendered as a numerical track. At control panel, go to "Statistiscs & analysis" → "Hypothesis test", to find its interface:

Define how many groups to compare

By default two groups are given. If you want to add more groups, click button :

Click button "Remove group" to remove last group.

Assign tracks to groups

Heatmap tracks can be assigned to a group by clicking button inside the group box:

Make selection from the heatmap selection panel to assign tracks to the group:

At least two non-empty groups are required to run hypothesis test.

More conveniently, you can right-click on track in genome heatmap or metadata color map and assign them to a group via menu option:

Please note that this option will only show up when the hypothesis test control panel is open.

Run hypothesis test

Kruskal-Wallis rank sum test is used, which is carried out by R function kruskal.test. Follow this link for details on this function.

One test is applied to each one of the data points. To adjust against multiple-testing effect, P-value correction could be applied through the drop-down menu. Again, this is carried out by R function p.adjust.

In the below example, testing was set on the ChIP-Seq tracks of two histone marks, (H3K27me3 in red as group 1, H3K36me3 in green as group 2) over some blood and immune cell samples from Roadmap Epigenomics Project:

Fdr-corrected P-values are plotted as downward-pointing bars. P-values are log10-transformed, the longer the bar is, the smaller the P-value or more significant the difference is. Gray bars indicate P-value of zero for the data points. A horizontal line is plotted to indicate where 0.05 the cutoff value is.

After modifying P-value correction option, you need to re-run hypothesis test to let it take effect.

Configure P-value track

Right click on P value track to show context menu:

Select wrench option to open configuration panel in floating toolbox:

To change P value plotting color click on the color blob on the top and select a color from the palette:

Change the cutoff value using drop-down menu, and color of the cutoff line via the second color blob. Following example shows the cutoff line was changed into gray, and drawn at 0.01:

Get regions with significant P values

You can get all regions (or data points) with P value below cutoff. Just select the option in the context menu:

A new window will be poped out showing data on regions with small-enough P-values:

Each row is one region, and each region corresponds to a data point (or column) in genome heatmap. Following the coordinate in the table are P-value, and track values over this region.

Under GROUP 1 and others are track names, and the cell values are data points used in hypothesis test.

Genome-wide hypothesis test

To view genome-wide hypothesis test result, click "Bird's eye view" option in menu:

The Browser will be running genome-wide hypothesis test. This involves a lot of computation and takes a short while. Please be patient and wait till the result appear in Bird's Eye View panel:

For further information, please refer to chapter of Bird's Eye View.


Last modifed: 09/29/2011