The basic steps for getting started with an analysis on the Research Analysis Platform (RAP) generally follow a three-step quickstart process, covering account setup, project configuration, and initial data access.
BEWARE - Work on this platform requires payment to be established.
Basic Overview¶
1. Create a UKB-RAP Account¶
Before accessing the RAP, you must:
Create a UK Biobank Access Management System (AMS) account and ensure you have received UK Biobank access approval from the Access Management Team (AMT).
Ensure you are listed as a collaborator on a UK Biobank-approved access application.
Once prerequisites are met, the steps for account creation and connection are:
Navigate to the Research Analysis Platform and click Create an Account (if you don’t have an existing DNAnexus account).
Fill out the Create New Account form, click Create Account, and activate the account via the emailed link.
Log into the Platform, complete your profile, and click Access Platform.
Connect your RAP account to your AMS account by accepting the Terms of Service and logging into the AMS when prompted. Your accounts are now connected, and you can use the Platform.
2. Create a Project and Dispense Data¶
All work on the RAP takes place within the context of a project. Projects allow a defined set of users to access specific datasets and conduct analyses.
On the RAP Projects screen, click the New Project button.
In the New Project wizard, enter a Project Name.
Enter the approved UK Biobank Application ID. Note that every project must be linked to one and only one access application.
Check the Dispense data to the project option to populate the project with data specified on the linked access application.
Choose a wallet for billing activities in the Billed To field.
Specify access permissions for copying, deleting, and downloading data in the Access section.
The process of dispensing data, due to the large size of the UK Biobank data, can take time—sometimes over an hour, or potentially between 20 minutes to 2 hours. You can monitor the status and percentage of populated data on the project list. Tabular data and linked health data are placed into a Spark SQL database and an associated dataset, while bulk data fields are dispensed as files within the /Bulk folder of your project.
3. Accessing Data and Starting Analysis¶
Once the data has been dispensed, you can start exploring and analyzing the data:
Accessing Datasets via the Cohort Browser¶
The Cohort Browser is a key tool for exploration, cohort definition, and visualization.
Access the dataset: Datasets must be prepared and ingested to be accessible via the Cohort Browser. To open a dataset, navigate to the project’s root folder and click the dataset (or select the dataset and click Explore Data).
Explore Data: In the Cohort Browser’s Overview tab, you see visualizations providing an introduction to the dataset. You can click the Add Tile button to explore what fields are available, which are organized in a folder structure inspired by the UK Biobank Showcase.
Define a Cohort: When you start exploration, an empty cohort is automatically created. You can narrow down your cohort by clicking the Add Filter button and selecting criteria based on data fields, genomic data, or join filters.
The Cohort Browser features can be used to further explore data, create charts (by adding data fields as tiles), or define and compare cohorts. Cohorts can be saved for later use or exported.
Advanced Analysis Options¶
Statistical Analysis using JupyterLab/Spark: Bioinformaticians can perform statistical analysis by creating JupyterLab environments backed by Spark clusters to directly query their data and create dataframes within a Python or R environment for further analysis.
Running Analysis Jobs: Once analysis jobs are run (e.g., using analysis apps or workflows), you can view their progress in the Monitor tab. This includes checking the status of jobs (running, errored out, successfully completed, or terminated).
Machine Learning (ML) Models: The RAP supports machine learning analysis and training models on imaging data, often leveraging Graphics Processing Units (GPUs). This is often done by selecting a JupyterLab session from tools and choosing a GPU instance type, such as
mem3 gpu. Several ML frameworks are available, including TensorFlow, PyTorch, or Scikit-learn.
Accessing Neuroimaging Data¶
Working with neuroimaging data collected by the UK Biobank (UKBB) on the Research Analysis Platform (RAP) requires ensuring that your project is correctly configured to access the large bulk data files associated with imaging and then utilizing the appropriate analysis environment.
Neuroimaging data—like magnetic resonance imaging (MRI), optical coherence tomography (OCT) imaging, or cardiac imaging—are considered bulk data-fields. The data dispensing process can take time, sometimes over an hour or between 20 minutes and 2 hours, depending on the scope of the application.
3. Locating and Accessing Neuroimaging Data Files¶
Once dispensing is complete, the data structure dictates how you access the imaging files:
Bulk Data Location: Bulk data files, including imaging data, are placed into the
/Bulkfolder within your project.Folder Structure: The
/Bulkfolder organizes files using subfolders based on the UK Biobank bulk field category. For instance, Brain MRI images will be found in folders within the Bulk directory. The folder structure includes categories and specific fields, such as/Bulk/Brain MRI/dMRI, which contains files from fields like Multiband diffusion brain images (DICOM #20218 and NIFTI #20250).Raw Data Formats: Typical raw imaging files you would work with include DICOM (often containing multiple 2D slices per folder, storing data and metadata like image resolution) and NIFTI (storing 3D voxel data, usually one file per volume, which is more compact and often preferable for machine learning applications).
4. Setting up the Analysis Environment¶
Depending on your analytical goal (working with pre-computed metrics versus running custom processing on raw images), you will need different environments:
| Analysis Goal | RAP Access Method/Environment Setup | Details |
|---|---|---|
| Working with Pre-computed Metrics | Cohort Browser | You can directly utilize Image Derived Phenotypes (IDPs) and other neuroimaging features that have been pre-computed by UKBB. These IDPs are accessible via the Cohort Browser, where you can load, filter, and analyze cohorts by filtering criteria such as volumes of gray matter. |
| Custom Processing/Statistical Analysis | JupyterLab Environments backed by Spark | Bioinformaticians can create JupyterLab environments backed by Spark clusters to query data and create dataframes in Python or R. From here, you can install and use open-source tools like FSL or FreeSurfer for neuroimaging analysis. Analysis steps might include DICOM to NIFTI conversion or brain extraction (school stripping) using tools like FSL’s bet command. |
| Machine Learning (ML) | JupyterLab with GPU Instance | To conduct machine learning analysis and train models on imaging data faster, you should select a GPU instance type for your JupyterLab session, such as mem3 gpu. ML frameworks like TensorFlow, PyTorch, or Scikit-learn are available. |