README.md

metadata

title: Polymer Discovery Platform
sdk: streamlit
python_version: '3.10'
app_file: Home.py

Polymer Discovery Platform

An integrated Streamlit platform for polymer screening and candidate discovery. The application combines property lookup, machine-learning prediction, molecular visualization, multi-objective discovery, AI-assisted query translation, novel polymer SMILES generation, and export to an automated molecular dynamics workflow.

What The Platform Does

The website is organized into eight modules:

• Property Probe: query a single polymer by SMILES or name and retrieve available database values with prediction fallback.
• Batch Prediction: run multi-property prediction for pasted, uploaded, or built-in polymer sets.
• Molecular View: render 2D and 3D molecular structures and export structure assets.
• Discovery (Manual): perform explicit constraint-based and multi-objective polymer screening.
• Discovery (AI): translate natural-language design requests into structured discovery settings with bring-your-own-key LLM support.
• Novel SMILES Generation: sample new polymer candidates with the pretrained RNN and filter against local datasets.
• Literature Search: search papers, stage evidence records, and review structured material-property extraction before promotion.
• Feedback: submit issue reports and feature requests through a webhook-backed form.

Core Capabilities

• Multi-source property lookup from EXP, MD, DFT, GC, and POLYINFO
• Property prediction across 28 polymer properties
• Large-scale screening over real and virtual candidate libraries
• Exact Pareto ranking with trust and diversity-aware selection
• AI-assisted prompt-to-spec generation for discovery workflows
• Novelty-filtered polymer SMILES generation
• Material-aware literature retrieval, evidence staging, and reviewer workflow
• ADEPT handoff for downstream molecular dynamics workflow packaging

Repository Layout

.
├── Home.py                      # Main Streamlit homepage
├── app.py                       # Compatibility entrypoint
├── pages/                       # User-facing application modules
├── src/                         # Prediction, discovery, lookup, and UI logic
├── literature/                  # Literature-mining pipeline components
├── scripts/                     # Utility and workflow scripts
├── data/                        # Lookup tables, discovery datasets, ADEPT files
├── models/                      # Trained prediction and generation assets
├── RNN/                         # Generator training/inference code
└── icons/                       # Application icons and branding assets

Data And Model Assets

This repository expects pretrained models and local data tables to be present. The application uses:

• source datasets such as EXP.csv, MD.csv, DFT.csv, GC.csv, POLYINFO.csv, and PI1M.csv
• derived property tables such as POLYINFO_PROPERTY.parquet and PI1M_PROPERTY.parquet
• trained checkpoint files under models/
• pretrained RNN assets under RNN/pretrained_model/ and models/rnn/pretrained_model/

If you clone only the code without the large assets, several app modules will not run correctly.

Local Development

Use Python 3.10.

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
streamlit run Home.py

Open http://localhost:8501.

Literature Dependencies

The production app now includes the literature workflow through requirements.txt. If you are working on the literature pipeline in isolation, you can still install:

pip install -r requirements-literature.txt

Environment Configuration

Create a local .env file if needed. The template is provided in .env.example.

Key variables used by the platform include:

LLM / Discovery AI

• CRC_OPENWEBUI_API_KEY
• OPENWEBUI_API_KEY
• OPENAI_API_KEY
• CRC_OPENWEBUI_BASE_URL
• OPENWEBUI_BASE_URL
• CRC_OPENWEBUI_MODEL
• OPENWEBUI_MODEL
• OPENAI_MODEL

The Discovery AI page also supports direct bring-your-own-key usage against supported providers from the UI.

Literature Pipeline

• PUBMED_EMAIL
• PUBMED_API_KEY
• SEMANTIC_SCHOLAR_API_KEY
• PAGEINDEX_API_KEY
• LITERATURE_MODEL_OPTIONS

Feedback / Analytics

• FEEDBACK_WEBHOOK_URL
• FEEDBACK_WEBHOOK_TOKEN
• APP_DEPLOYMENT_SOURCE

Running With Docker

docker build -t polymer-discovery .
docker run --rm -p 8501:8501 polymer-discovery

The container launches:

streamlit run Home.py --server.port=8501 --server.address=0.0.0.0 --server.headless=true

Notes For Deployment

• The app is designed as a Streamlit website.
• Heavy modules depend on local datasets and pretrained checkpoints being available at the expected paths.
• The AI-assisted discovery page requires a valid API key when using in-app LLM generation.
• The feedback page requires a configured webhook to receive submissions.

Citation And Use

If you use this platform in research or build on top of it, cite the associated paper once published. Until then, reference the repository and the MONSTER Lab platform description.