Cohort Browser

Overview

DNAnexus Apollo builds on the technological foundation of the core DNAnexus Platform to offer scientists and bioinformaticians an environment to store and query large sets of genomic, phenotypic, multi-omic, and other structured data. Researchers can bring their data to the Platform and leverage DNAnexus apps to ingest the data into queryable databases.

These databases can then be explored using the Cohort Browser. Scientists can filter the dataset by any data field and save these filtered samples as cohorts. These cohorts can be shared with other scientists and also can be used as inputs to analysis apps to perform such tasks as calculating allele frequencies or performing a GWAS analysis.

The Cohort Browser dashboard can show up to three tabs based on the configuration of the dataset: Overview, Data Preview, and either Genomics (if the dataset contains germline genomic data) or Somatic Variants (if the dataset contains somatic variant data). Tabs are loaded as the user clicks on them, so if there is no change in filtering, the tabs will stay cached and will not need to reload.

Bioinformaticians who wish to perform statistical analysis are able to create JupyterLab environments backed by Spark clusters to directly query their data and create dataframes within a Python or R environment for further analysis.

Accessing Datasets

Opening a Dataset Using the Cohort Browser

From the project where a dataset is located, go to Manage tab and select your dataset of interest. Click on the Explore Data action to open this dataset in Cohort Browser.

You can also access datasets via the Datasets page which is located under the Projects menu. The Datasets page displays all datasets you have access to, and enables you to browse and find a specific dataset without navigating through projects.

You can use the optional information panel to view further information about a selected dataset, including creator, sponsorship, and data provenance details.

Exploring Data in a Dataset

In the Cohort Browser's Overview tab, you'll see visualizations that provide an introduction to the dataset, and insights on the data it contains.

Adding a Data Field of Interest as a Tile

1. To create and view a chart visualizing data in a field, click the Add Tile button. The Add Tile dialog will open, showing a hierarchical view of all the data fields available in the dataset.

2. Browse the list or search an item by its title to narrow down the list.

3. Select a data field from the list. In the Data Field Details panel, you can see metadata on the selected data field, visualization preview, as well as options to customize chart types.

4. Confirm selection via the Add as Tile action. The new tile will appear on your dashboard.

Creating Multi-Variable Charts

Once you've selected a primary data field in the Add Tile dialog, you can add a secondary data field by clicking on the + icon next to an eligible secondary data field.

For certain chart types - such as Stacked Row Chart and Scatter Plot - you can re-order the primary and secondary data fields by dragging on the data field in the Data Field Details section.

Video Overview: Exploring New Datasets

This video provides a detailed overview of exploring new datasets using the Cohort Browser:

Defining a Cohort

When you start exploration on a dataset, an empty cohort is created automatically in the Cohort Browser. You can further narrow down your cohort by adding cohort filters. Cohorts can be saved and exported for later use.

Adding a Cohort Filter

1. From the cohort which you wish to edit, click on Add Filter button.

2. Select a data field you want to filter by, confirm by clicking on Add as Filter.

3. Select operators and enter values to filter by. Click on Apply Filter to confirm.

4. Filters added are displayed in corresponding cohort panels. You can edit a specific filter any time by clicking on it, which would bring up the Edit Filter dialog.

5. The default logical operator is 'AND'. To switch the operator to 'OR', click on the operator. For a filter group (a set of filters tied to 1 specific entity), all operators will be the same: all 'OR' or all 'AND'.

Once filters are added or edited, an updated cohort size will appear under name of the affected cohort. The dashboard will also auto-refresh to fetch updated results basing on latest cohort selection.

Adding a Genomic Filter

Germline Genomic Data

If your dataset includes germline genomic data, then you will have the option to add a genomic filter to your cohort.

1. From the cohort you wish to edit, click on Add Filter button.

2. Toggle to Geno tab.

3. Edit filter in Edit Genomic Filter dialog by one of the following criteria:

   1. Filter by genes and variant effects: Filter your dataset by variants of certain types and consequences within specified genes and/or genomic ranges. A maximum of 5 genes/ranges can be entered.

   2. Filter by a list of variant IDs. A maximum of 100 variants can be entered.

   3. If more than 1 range, gene, or variant is added, the values should be comma separated or each value must be on a new line.

4. Confirm edit by clicking on Apply Geno Filter button.

Similarly to the other cohort filters, a genomic filter is applied to the main entity of your dataset, such as patients or participants.

For datasets with canonical transcript information available, an additional toggle will appear in the Genomic Filter dialog titled "Match effects for canonical transcript only" which may be set to YES initially to restrict the results only to variants that have canonical information available.

Canonical transcripts, as defined by Ensembl, will be indicated with a blue marker next to their "Ensembl Transcript ID" in the Transcript column in the Genes/Transcripts table.

Somatic Variants Data

If your dataset includes data on somatic variants, the workflow for creating a genomic filter is similar to that for datasets containing germline data.

In the Add Filter / Edit Filter pane, you will see options enabling you to:

• Filter by genes and variant effects.

• Filter by a particular HGVS DNA or HGVS protein notation, preceded by a gene symbol.

• Filter by a list of variant IDs. A maximum of 10 variants can be entered.

For each somatic variant filter, you can specify if matching variants are to be used as inclusion criteria or exclusion criteria for your cohort. By default, you will be selecting patients who have at least one detected variant that matches the specified criteria. To select patients or participants who do not have any matching variant, click the "WITH" dropdown button and change its value to "WITHOUT".

You can create up to 10 somatic variant filters for each cohort.

Adding a Join Filter

When working with datasets that have multiple data entities, you can create a join filter by selecting data fields from a secondary entity and adding them as filters. An entity is a grouping of data around a unique item, event, or a concept. For example, a patient, visit, medication, or laboratory tests.

Join Filters are displayed as subrows deriving from the main entity. Depending on the entity to which your selected data field belongs, a join filter that reflects the relationship between those entities will be automatically created. To create a new cohort criteria using the join filters, click + Add filter or the Filter > Add filter on a tile. To add additional criteria to an existing criteria in a join click the Add additional criteria inline on the row of the chosen filter.

You can choose between the 'AND' or 'OR' logical operators when creating a cohort and comparing join filters. To switch between them, click on the logical operator. For a specific level of join filtering, joins are either all 'AND' or all 'OR'. Even when using 'OR' for two join filters, the implication that "this criteria exists" precedes the join level, that is, "where exists, join 1 or join 2".

Once a join filter is created, you can further define the secondary entity by adding additional criteria to the branch, or adding more layers of join filters deriving from the current branch. As you add more layers, the field selector automatically hides fields that are ineligible to be added based on the join.

For an example of interpreting join filters, consider the following:

The First Example cohort identifies all patients with a "high" or "medium" risk level who have a first (visit instance = 1) hospital visit and who also have had a lab test that was a "nasal swab". This lab test does not necessarily have to be conducted at the time of the patient's first hospital visit. In the Second Example, the cohort includes all patients with a "high" or "medium" risk level who had the "nasal swab" test performed on the first visit.

Video Overviews: Cohort Definition

Define Your Cohort

Build Cohort Definitions Across Complex Datasets

Adding Genomic Based Cohort Definitions

Setting Up a Cohort Browser Dashboard

This video provides an overview of setting up your dashboard as part of defining and refining a cohort:

Combining Cohorts

You can create complex cohorts by combining existing cohorts from the same dataset. Cohort combine can be accessed via the "Compare / Combine Cohorts" menu located at the top of the page.

You can also create a combined cohort basing on the cohorts already being compared.

The Cohort Browser supports the following combination logic:

Once a combined cohort is created, you can inspect the combination logic and its original cohorts in the cohort filters section.

Cohort Table

In the Data Preview tab, the Cohort Table shows records that are within your current cohort selection split up by the entity they are on. You can add or remove data fields as columns via the column customization menu, which is located in the top-right corner of the table. As you add fields, entities are automatically split out. You can have up to 5 different entities showing at a time. In the entity drop down, you can toggle between specific entities you've added and remove them outright.

Click on table column headers to access more functionalities including sorting and searching in a specific column and the data field information. From the data field information, you can quickly add the field as a tile or a filter if it has not been added yet.

Export Cohort Table

You can export table information either as a list of record IDs or a CSV file. Export options are available on the top-right corner of the table once you have selected multiple table rows.

Variant Browser

The Variant Browser shows variants that are present in current cohort selection.

For datasets containing germline data, the Variant Browser appears in the Genomics tab. For datasets containing somatic variants data, it appears in the Somatic Variants tab.

Germline Data

For datasets containing germline data, the Variant Browser includes a lollipop plot displaying allele frequencies for variants in a specified genomic region.

The table below the lollipop plot displays a list of the same variants in tabular format, along with further annotation information including:

• Type: whether the variant is a SNP, deletion, insertion, or mixed.

• Consequences: The impact of variant according to SNPEff. For variants with multiple gene annotations, this column displays the most severe consequence per gene.

• Population Allele Frequency: Allele frequency calculated across entire dataset from which the cohort is created.

• Cohort Allele Frequency: Allele frequency calculated across current cohort selection.

• GnomAD Allele Frequency: Allele frequency of the specified allele from the public dataset GnomAD.

If canonical transcript information is available, the following three columns with additional annotation information will appear in the Allele Table:

• Consequences (Canonical Transcript): Canonical effects per each associated gene, according to SnpEff.

• HGVS DNA (Canonical Transcript): HGVS (DNA) standard terminology per each associated gene with this variant

• HGVS Protein (Canonical Transcript): HGVS (Protein) standard terminology per each associated gene with this variant

To view further annotation information, you can go to the detail page of a given variant by clicking on the link in the Location column .

Somatic Variant Data

Lollipop Plot

For datasets containing somatic variant data, the Variant Browser includes a lollipop plot showing somatic variants for the canonical protein of a single gene, that occur in the cohort under examination. Interpreting and working with this chart is similar to working with the lollipop plot included in the Variant Browser for datasets containing germline data.

Consider the following points:

• The lollipop plot displays variant information in one gene / canonical protein at a time.

• Each "lollipop" represents amino acid changes at a given location such as "Thr322Ala". Location information is visualized as horizontal position (X axis) and affected sample frequency in the current cohort is visualized as height (Y axis).

• Each lollipop is color-coded by consequence according to the canonical transcript. Lollipops that cover more than one consequence types are color-coded as "Multiple Consequences"

You can inspect variant statistics under each specific consequence, by interacting with the legend panel and selecting one color at a time. Selecting a particular lollipop in the plot will apply a filter to the variants table, such that only variants corresponding the selected lollipop are displayed.

You can navigate to a different gene by entering the gene symbol in the Go to Gene field. This will also update the variants table by automatically navigating to the corresponding genomic region.

Variant Frequency Matrix

For datasets containing somatic variant data, the Variant Browser includes a chart illustrating the overall variant landscape across the top-mutated genes, for the current cohort.

Consider the following points:

• The genes are sorted in descending order of percent of affected samples.

• Samples are displayed, from left to right, from those that have the greatest number of mutated genes across all genes, to those that have the least.

• A sample is considered affected, if the sample has at least one detected mutation of high or moderate impact within the canonical transcript for that gene.

• Samples are color-coded by consequences. Samples with two or more detected variants are color-coded as "Multi Hit".

• Only consequences of high or moderate impact (as defined by Ensembl VEP version 109) are included in this visualization.

• The variant frequency matrix plot displays up to 50 top mutated genes, for up to 500 samples for any given cohort.

You can inspect variant statistics under each specific consequence, by interacting with the legend panel and selecting one color at a time.

Hovering over a particular sample cell will open a popover window showing detailed information on the sample, including:

• The Sample ID

• The count of variants per consequence, within the respective gene

Variants Table

For datasets containing somatic variant data, the Variant Browser includes a Variants Table that provides, in tabular format, details on the variants found in a particular genomic region, in samples for the current cohort.

The table displays detail on the same genomic region as the lollipop plot.

Information displayed in the Variants Table includes:

• Location of variant

• Reference allele of variant

• Alternate allele of variant

• Type of variant

• Variant consequences, with entries color-coded by level of severity

• HGVS cDNA

• HGVS Protein

Exporting the Variant Table

You can export selected variants in the table as a list of variant IDs or a CSV file. Export options will appear at the top-right corner of the table once you have items selected.

Save Cohort

You can save your cohort selection to a project as a Class: Record object by clicking the Save icon in the top-right corner of the cohort panel.

Cohorts will be saved with the filters applied, along with the latest set of visualizations and dashboard layout information. Similar to Dataset objects, Cohort objects can be found under the Manage tab in your selected projects, and can be re-opened via the Explore Data option.

Export Cohort

You can export a list of main entity IDs in your current cohort selection as a CSV file. This action can be found next to the Save Cohort button, on the top-right corner of cohort panel.

Dashboard Views

Dashboard views contain layout and configuration information that can be re-used during cohort browsing. You can save or load a dashboard view via the Dashboard Actions menu located at the top-right corner of the header area. Dashboard views are saved as "Type: DashboardView" objects, which once saved also show up in selected project folders.

Comparing Cohorts

You can compare two cohorts by adding both cohorts into the Cohort Browser. In cohort compare mode, all visualizations are converted to show data from both cohorts.

The Compare Cohort action can be found in the header area next to the cohort title. You can create a new cohort, duplicate the current cohort, or load a previously saved cohort.

In compare mode, you can continue to edit both cohorts and visualize the results dynamically.

You can compare a cohort with its complement in the dataset by selecting the "Not In …" option in the Compare / Combine menu. Similar to combining cohorts, you must first save your current cohort before creating its not-in counterpart.

Video Overview: Comparing Cohorts During Cohort Definition

Download Restrictions

The Cohort Browser adheres to a project's download policy restrictions.

1. If a database is in a project that has a restricted downloadPolicy, then a Cohort Browser that shows a dataset, cohorts, or dashboards pointing to that database should not allow downloads, regardless of project.

2. If all copies of the dataset are in a project that has a restricted downloadPolicy, then a Cohort Browser that shows a dataset, cohorts, or dashboards pointing to that dataset should not allow downloads, regardless of project.

   1. Check all copies of the dataset.

   2. If at least one copy is in a "download allowed" then the download will be allowed.

3. If the cohort or dashboard you are launching has a restricted downloadPolicy, then a Cohort Browser that shows the cohort/dashboard should not allow downloads.

Create Cohorts via CLI

The dx command create_cohort generates a new Cohort object on the platform from an existing Dataset or Cohort object and using a list of primary IDs. The filters are applied to the global primary key of the dataset/cohort object. When the input is a CohortBrowser record, the existing filters are preserved and the output record has additional filters on the global primary key. The filters are combined in a way such that the resulting record is an intersection of the IDs present in the original input and the IDs passed through CLI. For additional details, see create_cohort and example notebooks in the public GitHub repository, DNAnexus/OpenBio.