MolVE: An Open-Source Web Platform for Visualizing and Evaluating AI-Designed Molecules to Aid in Prioritization

This repository contains the code used to generate the results reported in the paper: MolVE: An Open-Source Web Platform for Visualizing and Evaluating AI-Designed Molecules to Aid in Prioritization.

@article{doi:10.1021/acs.jcim.5c02412,
  author  = {Rigoni, Davide and Sperduti, Alessandro and Moro, Stefano},
  title   = {MolVE: An Open-Source Web Platform for Visualizing and Evaluating AI-Designed Molecules to Aid in Prioritization},
  journal = {Journal of Chemical Information and Modeling},
  volume  = {0},
  number  = {0},
  pages   = {null},
  year    = {0},
  doi     = {10.1021/acs.jcim.5c02412},
  note    = {PMID: 41667067},
  URL     = {https://doi.org/10.1021/acs.jcim.5c02412},
  eprint  = {https://doi.org/10.1021/acs.jcim.5c02412}
}

MolVE is an open-source, web-based platform for collaborative, expert-driven assessment of molecules (e.g. AI-generated candidates). It provides secure authentication, dataset management (SDF import/export), a simple evaluation workflow (one molecule at a time), interactive 2D/3D visualisation (ChemDoodle + JSmol), and programmatic APIs (API tokens) for automation and integration.

The system is shipped as a Dockerised stack:

• Web app: React/Vite frontend + Node/Express backend (TypeScript)
• Database: PostgreSQL
• Optional microservice: Python/FastAPI + RDKit for molecular processing (SMILES↔SDF, property extraction, predictions)

If you are new to the platform, the in-app guide is the best starting point: open /guide once the app is running.

Dependencies

Recommended (supported) way to run MolVE:

• Docker + Docker Compose

Structure

This repository is organised as a small monorepo:

• docker-compose.yaml — development stack (bind-mounts source, exposes debug ports)
• docker-compose.prod.yaml — production-like stack (immutable images; HTTPS preferred)
• services/app/ — TypeScript full-stack web app
  • client/ — React frontend (pages, components, Tailwind)
  • server/ — Express backend (API routes, auth, DB access)
  • shared/ — shared schemas/types (frontend + backend)
  • initial_configs/ — container startup scripts
• services/python-service/ — FastAPI/RDKit microservice
• services/db/ — Postgres volume location and init SQL
• tests_tools_to_use/ — small files/scripts useful for local testing (e.g. SDF samples, Python API script)

Usage

Quick start (development)

Build and run the full stack:

docker compose -f docker-compose.yaml up --build

Then open:

• Web app (HTTP): http://localhost:5000
• Python service (dev only): http://localhost:8000

Default development ports exposed by docker-compose.yaml:

• 5000 web app (HTTP)
• 9229 Node.js inspector (only if you attach a debugger)
• 8000 Python service (debug/development)
• 5432 PostgreSQL (debug/development)

Stop the stack:

docker compose -f docker-compose.yaml down

First login

On startup, the backend will create an initial admin user if it does not exist yet:

• username: admin
• password: admin

Log in at /login, then open /admin-dashboard to create evaluator accounts, upload molecules, and configure evaluation settings.

Production-like run

Run using the production compose file:

docker compose -f docker-compose.prod.yaml up --build

In production mode the app prefers HTTPS on port 443. A self-signed certificate is generated automatically inside the container if missing.

If you don’t want HTTPS, you can remove the 443:443 port mapping in docker-compose.prod.yaml or provide your own certificate paths via SSL_CERT_PATH / SSL_KEY_PATH.

Common workflows

• Upload molecules: use the admin dashboard to upload SDF files (the backend will call the Python/RDKit service to parse/process them).
• Evaluate molecules: evaluators typically use /user-evaluation to rate one molecule at a time.
• Export results: admins can download molecules/evaluations as CSV/SDF from the admin dashboard.

API tokens and programmatic access

MolVE supports per-user API tokens for programmatic access. Create tokens from the UI, then call versioned endpoints under /api/v1/... with:

Authorization: Bearer <token>

Some endpoints are restricted to admin-owned tokens (for example, bulk import endpoints). Prediction is available via POST /api/v1/predict and the stable alias POST /api/predict.

For a working Python example, see tests_tools_to_use/test_python_api.py.

Notes

• Upload limits: the maximum SDF upload size is controlled by 100MB (see the compose files).
• MolVE was tested on a low-resource AWS EC2 t3.micro instance located in Stockholm, equipped with 2 vCPUs and 1 GB of RAM. Under this setup, MolVE effectively handled datasets suitable for human evaluation. Specifically, it has been tested with datasets containing up to 10,000 molecules without issues. The system was configured to accept SDF files up to 100MB in size, with the capability to upload them multiple times. While uploading large files may take additional time, primarily due to variations in internet connection speed, molecule processing itself is efficiently expedited by directly parsing the SDF file as text, eliminating the need for additional packages. The ML model's prioritization score is computed in the background within MolVE's backend. Initially, the ML predictions are displayed as ``Pending,'' but they are updated in real-time once calculated, allowing users to continue their workflow without interruption.

License

Apache License, Version 2.0