← Back to Home

Data Sources and Tests: What types of data can be visualized

There can be multiple possible sources/target objects defined for each visualization, For example, a visualization may be applicable to both SAM and BAM files and have a data_source defined for each:

            <test type="isinstance" test_attr="datatype" result_type="datatype">tabular.Sam</test>
            <test type="isinstance" test_attr="datatype" result_type="datatype">binary.Bam</test>

When the registry is deciding whether to render a link:

  • Each data_source is checked in turn.
  • If the source of data passes any of the tests in a data_source a link is rendered.
  • If the data_source tests fail, the next data_source is checked and the process is repeated until there are no more data_sources.

In the example above, if the registry was checking a BAM file, the first data_source's tests would fail, the registry would check the next data_source, those tests would pass, and a link would be rendered.

This control flow described above implicitly means that it's better to order data_sources from more specific first to more general after. This allows us to send different or added settings based on the 'cascading' effect of the flow.


Inside each data_source, the first test uses the data_source sub-element model_class. If the target object is the same python class as the class listed here, then the remaining data_source tests are then run.


model_class is the only required test for all data_sources and must be defined.

In the majority of cases, your visualization will be using data from a dataset in a user's history - so HistoryDatasetAssociation will be the proper setting.


test elements define tests to see if a visualization can be applied using the model as a source of data. If any single test passes, the visualization is considered applicable and a link will be rendered (in other words, the tests are effectively OR'ed). There can be zero or more data_source tests (sometimes just checking the model_class alone is enough information).

(Note: the model_class is an exception to tests being OR'd: it is effectively AND'd to the whole set of tests.)

Test types are currently limited to:

  1. isinstance: testing a LibraryDatasetDatasetAssociation or HistoryDatasetAssociation's .datatype attribute for class inheritance:
<test type="isinstance" test_attr="datatype" result_type="datatype">data.Newick</test>
  1. has_dataprovider: testing if the attribute of a data_source/model has a specific DataProvider:
<test type="has_dataprovider" test_attr="datatype">node-edge</test>

or 3) eq: using string comparison of any model's attribute (the default - no types are required):

<test test_attr="dbkey">hg18</test>

Note that, for datatype testing:

  • you need to use the datatype module namespace (e.g. data.Newick or tabular.Tabular)
  • the use of isinstance allows you some flexibility for your testing, as any sub-class of a datatype will still pass the datatype test (e.g. a SAM file will be considered applicable if you test for the datatype tabular.Tabular)
  • datatypes are checked against the datatype registry of your Galaxy installation

Each data_source can define zero or more to_param elements. Each to_param will add one key/value pair to the query string of the URL of your link allowing control over what information is contained in your visualization link and controlling what is passed to your visualization code. For example:

        <test type="isinstance" test_attr="datatype" result_type="datatype">tabular.Sam</test>
        <to_param param_attr="id">dataset_id</to_param>

In this case, the to_param will take the attribute of the target object 'id' (param_attr="id") and attach a key value pair on the generated link query string of: ?dataset_id=<the HDAs id>.

Note: ids are automatically encoded when the link is generated.

The final link generated points to the exposed render method of the visualization.py controller:

<a href="http://localhost:8080/visualization/show/myvis?dataset_id=f7bb1edd6b95db62">My Visualization</a>

The visualization name is passed as the first argument and appears after the show 'verb'.

Other values can be passed in as well depending on the data source either using the target's attributes or direct assignment:

  <test type="isinstance" test_attr="datatype" result_type="datatype">data.Data</test>
  <to_param param_attr="id">dataset_id</to_param>
  <to_param assign="ldda">hda_ldda</to_param>
<a href="http://localhost:8080/visualization/show/myvis?dataset_id=f7bb1edd6b95db62&hda_ldda=ldda">My Visualization</a>

When the link generated from the data_source is used, resources listed in params are parsed from the query string and sent to the template. 'Resources' can be considered any data that needs to be parsed or converted from a string before being sent to the visualization template (primitives such as ints or booleans, more complex data such as models, or lists of either).

The registry uses the ResourceParser class located in lib/galaxy/visualization/plugins/resource_parser.py to do this.

For example, the scatterplot visualization has the following params (it only needs one 'resource' - the HDA target):

      <param type="dataset" var_name_in_template="hda" required="true">dataset_id</param>
  • The text node of the element is the key of the key/value pair in the query string: dataset_id
  • The required="true" attribute tells the registry to throw an error immediately if the dataset_id key is not in the query string of the incoming link
  • The type tells the registry to convert that value into a dataset (implicitly the dataset type receives an id)
  • The registry will attempt to get the full model of the dataset (SQLAlchemy eagerloading is on)
  • The var_name_in_template tells the registry to pass the dataset model to the template using the context variable name hda

Other complex params are also parsed:

  • type="hda_or_ldda": for visualizations that accept either an HDA or LDDA
  • type="dbkey": for visualizations that require the genome build of a target object
  • type="json": when data is bundled and passed as JSON (the string will be parsed into a python structure for the template)

Primitive types are parsed as well: 'str', 'bool', 'int', 'float' ('str' is the default type).

Besides the param attributes above, the following are also available:

  • csv : csv="true" will split the query string value at commas, and recursively parse each element using the type attribute sending the final list to the template. For example: <param type="int" csv="true">indeces</param> would parse the query string key/value pair ?indeces=3,4,9,12& and send it to the template as: indeces = [ 3, 4, 9, 12 ]
  • default : you can provide a default value for a param and, if it's missing from the query string, that value will be parsed using type and sent to the template

Any parameters defined in the params section of your config file are available here. In addition, other variables are available (see VisualizationsRegistry/Code for more info).

Other configuration options

You can also optionally specify a render_location element. This becomes the target attribute of the link generated for your visualizations. If you want the visualization to load in a new window, set this to _blank, in the top frame (and therefore remove the two side panels) use _top, or if you're fine loading it in the center panel with both side panels in place use galaxy_main (this is the default render_location).


You can test the validity of your XML registry file by using the DTD defined in visualizations_conf.xml itself. On the command line:

xmllint --valid --noout myvis.xml

If there's an error parsing the configuration of a visualization, you'll find an error similar to this in the server logs:

galaxy.web.base.pluginframework WARNING 2014-01-28 10:26:04,147 VisualizationsRegistry, plugin load failed or disabled:
./config/plugins/visualizations/sweepster. Skipping...

If a configuration loads successfully and you're not seeing a link to it on a dataset or data source that should display one:

  • Check the configurations model_class - it is case sensitive and the class should be part of model/__init__.py
  • Check the order of your data_sources - order matters here - go from specific first to general
  • Check your tests - attributes must be valid attributes of the model_class

If your link isn't turning out properly, check your param definitions. Again, they need to be valid attributes of the model_class.