Self-Driving Discovery of Electric Vehicle Coolants

This repository contains the implementation of the computational screening framework presented in:

Self-Driving Discovery of Electric Vehicle Coolants

The framework enables autonomous identification of high-performance coolant candidates through a modular, multi-stage pipeline integrating data acquisition, machine learning property prediction, mixture behavior modeling, structural filtering, and multi-criteria decision analysis (VIKOR).

---

🔬 Workflow Overview

The discovery pipeline follows a structured screening funnel:

1. PubChem Acquisition
   Automated retrieval of molecular structures and identifiers.

2. SMILES Validation & Structural Filtering

   • Removal of invalid SMILES\
   • Substructure screening (e.g., ester detection)

3. Thermophysical Property Prediction (ML Models)
   Machine learning models trained to predict:

   • Density\
   • Heat capacity\
   • Boiling point\
   • Melting point\
   • Corrosion\

4. Dynamic Viscosity Prediction
   Ensemble-based model for viscosity estimation.

5. Mixture Non-Linearity & Excess Property Screening
   Identification of favorable non-ideal mixture behavior.

6. VIKOR Multi-Criteria Ranking
   Final prioritization using weighted multi-objective optimization.

---

📁 Repository Structure

data/                # Input and processed datasets
notebooks/           # Jupyter notebooks for experiments and analysis
src/filters/         # Modular screening and filtering modules
requirements.txt     # Python dependencies
README.md

---

⚙️ Installation

Recommended (Conda Environment)

conda create -n evcoolants python=3.10
conda activate evcoolants
conda install -c conda-forge rdkit
pip install -r requirements.txt

---

📜 License

---

---

📖 Citation

If you use this code, please cite:

Self-Driving Discovery of Electric Vehicle Coolants
<Journal Name, Year>
<DOI>{=html}

---

👤 Author

Mahyar Rajabi Kochi
AI4ChemS - University of Toronto