Water-in-Salt Zn Electrolyte Atlas

Interactive research companion for the ICEMEE 2026 manuscript: Machine-learning analysis of organic co-solvents for water-in-salt zinc-ion battery electrolytes.

This repository packages the curated dataset, site data-generation pipeline, and a static web app for inspecting model outcomes, SHAP-based interpretation, and Pareto solvent recommendations.

Project scope

The atlas is designed to make cross-study comparison explicit and reproducible:

• Dataset curation across reported formulations and solvent descriptors
• Model result reporting for cycle lifespan and ionic conductivity (plus negative controls for CE and ESW)
• Interpretability surfaces using SHAP-derived feature importance summaries
• Decision support views with multi-objective (lifespan vs. conductivity) candidate ranking

The current generated payload reports:

• 108 formulations
• 63 source papers
• 23 engineered features
• 29 tracked co-solvents

Repository layout

.
├── data/
│   ├── formulations_curated.csv
│   ├── solvent_descriptors.csv
│   └── source_registry.csv
├── public/
│   └── paper/                               # manuscript PDF
├── scripts/
│   └── build_site_data.py                   # CSV -> src/data/generated/siteData.json
├── src/
│   ├── components/                          # dataset/model/shap/pareto/reference panels
│   ├── data/generated/siteData.json         # generated app payload
│   └── App.tsx
└── .github/workflows/deploy-pages.yml       # GitHub Pages build and deploy

Data and methodology summary

Input datasets

1. data/formulations_curated.csv
   Row-level electrolyte formulations with performance fields (e.g., lifespan, conductivity, CE, ESW), chemistry descriptors, and extraction provenance.
2. data/source_registry.csv
   Bibliographic/source registry used by the references panel.
3. data/solvent_descriptors.csv
   Solvent-level descriptor table (DN, AN, LogP, TPSA, viscosity, dielectric constant, boiling point, etc.).

Site data generation

scripts/build_site_data.py composes a single JSON payload consumed by the frontend (src/data/generated/siteData.json).
The script aggregates:

• Year/salt/status distributions
• Feature group definitions (23 total)
• Benchmark model summary tables
• SHAP feature summaries
• Pareto candidate lists and recommendation text
• Formulation/source/descriptor records for interactive views

Development

Requirements

• Node.js 22 (aligned with CI workflow)
• npm
• Python 3.10+ (for scripts/build_site_data.py)

Local setup

npm install
python scripts/build_site_data.py
npm run dev

Production build

python scripts/build_site_data.py
npm run build
npm run preview

Deployment

GitHub Pages deployment is defined in .github/workflows/deploy-pages.yml and triggers on pushes to main.
The workflow installs dependencies, regenerates siteData.json, builds the Vite app, uploads dist/, and deploys via actions/deploy-pages.

Reproducibility and limitations

• The site intentionally keeps cross-paper transfer realism as a central constraint.
• Current signals are strongest for lifespan and ionic conductivity; CE and ESW are included for transparency but remain weakly learnable in the present corpus.
• Recommendation outputs are intended as screening guidance, not standalone experimental validation.