🫀 iCARE4CVD Cohort Explorer

Webapp enabling to upload and explore cohorts metadata, built for the iCARE4CVD project.

It interacts with a privacy computing platform (Decentriq) to create secure workspace where data scientists can run analysis on the selected cohorts. The cohorts data is uploaded only to Decentriq, the explorer only uses cohorts metadata.

It aims to enable data custodians and data scientists to:

• 🔐 Login with their Decentriq account (OAuth based authentication, can be easily switch to other providers). Only accounts with the required permissions will be able to access the webapp.
  • ✉️ Contact Decentriq to request an account if you are part of the iCARE4CVD project
• 📤 Data custodians upload CSV cohort metadata files describing the variables of a study cohort
• 🔎 Data scientists explore available cohorts and their variables through a web app:
  • Full text search across all cohorts and variables
  • Filter cohorts per types and providers
  • Filter variables per OMOP domain, data type, categorical or not
• 🔗 Data custodians can map each variable of their cohorts to standard concepts, sourced from OHDSI Athena API (SNOMEDCT, LOINC...) through the web app.
  • Mapping variables will help with data processing and exploration (⚠️ work in progress)
  • We use namespaces from the Bioregistry to convert concepts CURIEs to URIs.
• 🛒 Data scientists can add the cohorts they need to perform their analysis to a Data Clean Room (DCR) on the Decentriq platform.
  • Once complete, the data scientists can publish their DCR to Decentriq in one click.
  • The DCR will be automatically created with a data schema corresponding to the selected cohorts, generated from the metadata provided by the data custodians.
  • The data scientist can then access their DCR in Decentriq, write the code for their analysis, and request computation of this code on the provisioned cohorts.

Important

Only the owner of the cohort (as described in the spreadsheet holding all cohorts generic metadata), and the platform admins, can upload the data dictionary or edit mappings for a cohort.

Note

You can reupload a cohort dictionary that have been already uploaded (in case you need to fix something). The mappings defined via the Cohort Explorer will be kept, as long as the variables names do not change.

⚠️ Known issues

Here are a known "issues" with the Cohort Explorer, and how to fix them:

• After a period of inactivity you might see a black screen with an error message, in this case just reload the page

🗺️ Technical overview

This platform is composed of 3 main components:

• Oxigraph triplestore database containing the cohorts and their variables metadata, exposing a SPARQL endpoint only available to the backend API.

  • The data stored in the triplestore complies with the custom iCARE4CVD OWL ontology. It contains 3 classes: Cohort, Variable, and Variable category. You can explore the ontology classes and properties here.
  • Oxigraph has not yet reached release 1.0, but it is already stable enough for our currently expected use. It has the advantages of being open source (important for accountability and trust), and developed in Europe. If missing features appears to be blocking, consider migrating to OpenLink Virtuoso, you'll only need to update the function that upload a RDFLib graph as file.

• backend/ server, built with python, FastAPI and RDFLib.

• frontend/ web app running in the client browser, built with TypeScript, NextJS, ReactJS, TailwindCSS, and DaisyUI.

🐳 Everything is deployed in docker containers define in the docker-compose.yml files.

🔐 Authentication is done through the Decentriq OAuth provider, but it could be replaced by any other OAuth provider easily. Once the user logged in through the external OAuth provider, the backend generates an encrypted JWT token, which is passed to the frontend using HTTP-only cookies.

Note

All metadata about cohorts and variables are retrieved by one mighty SPARQL query, and passed to the frontend as one big dictionary. Filtering and searching is then done in TypeScript on this cohorts dictionary.

We expect the amount of metadata for all cohorts will stay small enough to be handled directly on the client. If it becomes too big, it can be replaced by performing search and applying filters using SPARQL queries, to only retrieve metadata about relevant cohorts.

☑️ To do

• Integrate the LUCE blockchain component for data sharing consent:
  • We will store blockchain addresses, handle authentication, and add the UI elements directly in the Cohort Explorer (we can even store private keys or do wallet stuff there too if needed)
  • But we need to be able to query the blockchain easily through an API from our system (a basic HTTP OpenAPI would suffice, e.g. built with FastAPI)

🧑‍💻 Development

Warning

For deploying the Cohort Explorer in development or production you will need the spreadsheet containing the general cohorts informations. It needs to be an excel spreadsheet named iCARE4CVD_Cohorts.xlsx with a sheet named Descriptions with the following columns:

Name of Study	Type	Institution	Contact partner	Email	Study type	N	Start date	End date	Study duration	Ongoing	Patient population	Primary objective	Primary endpoints	Secondary endpoints

📥 Install dependencies

1. Install hatch for managing python projects, and pnpm for TS/JS projects

   pip install hatch

2. Create a backend/.env file with secret configuration:

   AUTH_ENDPOINT=https://auth0.com
   CLIENT_ID=AAA
   CLIENT_SECRET=BBB
   DECENTRIQ_EMAIL=ccc@ddd.com
   DECENTRIQ_TOKEN=EEE
   JWT_SECRET=vCitcsPBwH4BMCwEqlO1aHJSIn--usrcyxPPRbeYdHM
   ADMINS=admin1@email.com,admin2@email.com

3. Put the excel spreadsheet with all cohorts metadata in data/iCARE4CVD_Cohorts.xlsx. Uploaded cohorts will go to separated folders in data/cohorts/

Warning

There is a bug with pandas when conditional cells are used in the excel spreadsheet. To remove conditional cells copy the whole sheet content, delete the current content, and paste the original sheet content without formatting (ctrl+shift+v)

Important

For the authentication to the Decentriq OAuth provider to work you need to deploy the backend on http://localhost:3000

⚡ Start for development

In development it is more convenient to start all components like the database with docker, and the backend/frontend outside of docker.

Start the database with docker:

docker compose up db

In a different terminal, start the backend:

cd backend
hatch run dev

In another terminal, start the frontend:

cd frontend
pnpm dev

Tip

Alternatively you can start the whole stack in development with docker compose in 1 command, but you won't get hot reload for the frontend (you will need to rebuild the frontend image for changes to be taken into account):

docker compose up

Note

In development the user requesting a DCR will be added to as data owner of all cohorts dataset requested for development purpose (so they can provision the data themselves, and to avoid spamming emails owners when developing)

🧹 Code formatting and linting

Automatically format Python code with ruff and black, and TypeScript code with prettier:

./scripts/fmt.sh

🆙 Upgrading dependencies

For the NextJS frontend:

cd frontend
pnpm up --latest

🐳 Deploy

Deploy on a server in production with docker compose.

Put the excel spreadsheet with all cohorts metadata in data/iCARE4CVD_Cohorts.xlsx. Uploaded cohorts will go to separated folders in data/cohorts/

1. Generate a secret key used to encode/decode JWT token for a secure authentication system:

   python -c "import secrets ; print(secrets.token_urlsafe(32))"

2. Create a .env file with secret configuration:

   AUTH_ENDPOINT=https://auth0.com
   CLIENT_ID=AAA
   CLIENT_SECRET=BBB
   DECENTRIQ_EMAIL=ccc@ddd.com
   DECENTRIQ_TOKEN=EEE
   JWT_SECRET=vCitcsPBwH4BMCwEqlO1aHJSIn--usrcyxPPRbeYdHM
   ADMINS=admin1@email.com,admin2@email.com

3. Deploy the stack for production:

   docker compose -f docker-compose.prod.yml up -d

We currently use nginx-proxy for routing through environment variables in the docker-compose.yml file, you can change for the proxy of your liking.

🪄 Database administration

🗑️ Reset database

Reset the database by deleting the data/db folder:

rm -rf data/db

Next restart of the application the database will be re-populated using the data dictionaries CSV files stored on the server.

Warning

Resetting the database only if really necessary, it will cause to lose:

• All concept mappings added from the Cohort Explorer
• The info about Decentriq airlock data preview for cohorts that have been uploaded (it will default to false when recreating the database, admins can update them by downloading and reuploading the cohorts with the right airlock setting)

💾 Backup database

It can be convenient to dump the content of the triplestore database to create a backup.

To backup the triplestore in development:

curl -X GET -H 'Accept: application/n-quads' http://localhost:7878/store > data/triplestore_dump_$(date +%Y%m%d).nq

In production you will need to run it in the backend docker container, e.g. with:

docker compose exec backend curl -X GET -H 'Accept: application/n-quads' http://db:7878/store > data/triplestore_dump_$(date +%Y%m%d).nq

Caution

The path given for triplestore_dump.nq is outside the docker container

♻️ Reload database from backup

You can easily reload the dump in an empty triplestore.

Change the name of the backup file to yours, with the right date:

curl -X POST -T data/triplestore_dump_20240225.nq -H 'Content-Type: application/n-quads' http://localhost:7878/store

In production you will need to run it in the backend docker container, e.g. with:

docker compose exec backend curl -X POST -T /data/triplestore_dump_20240225.nq -H 'Content-Type: application/n-quads' http://localhost:7878/store

Caution

The path given for triplestore_dump.nq is inside the docker container

🚚 Move the app

If you need to move the app to a different server, just copy the whole data/ folder.

✨ Automatically generate variables metadata

Experimental: you can use the csvw-ontomap python package to automatically generate a CSV metadata file for your data file, with the format expected by iCARE4CVD. It will automatically fill the following columns: var name, var type, categorical, min, max. But it does not properly extract datetime data types.

Install the package:

pip install git+https://github.com/vemonet/csvw-ontomap.git

Run profiling, supports .csv, .xlsx, .sav:

csvw-ontomap data/COHORT_data.sav -o data/COHORT_datadictionary.csv