/MIML-Project_Database-for-Polymer-Informatics

Primary LanguageHTMLMIT LicenseMIT

MLMI: Machine Learning-Assisted Exploration of Thermally Conductive Polymers

This project is built as part of an academic exploration into machine learning-driven materials informatics. It uses publicly available data from the PI1M benchmark database to analyze and predict the thermal conductivity of polymer materials using high-throughput molecular dynamics simulation data and ML models.

📘 Overview

In recent years, machine learning-assisted molecular dynamics (ML-MD) has emerged as a powerful approach to accelerate materials discovery.
This project reproduces and extends concepts from the research article:

Machine Learning-Assisted Exploration of Thermally Conductive Polymers Based on High-Throughput Molecular Dynamics Simulations
Ruimin Ma and Tengfei Luo, Journal of Chemical Information and Modeling, 2021.

We utilize the PI1M dataset, a large-scale benchmark for polymer informatics, to train and test models for predicting polymer properties such as:

  • Density
  • Glass transition temperature (Tg)
  • Melting temperature (Tm)
  • Dielectric constant
  • Synthetic accessibility (SA) score

🧠 Objective

The goal of this project is to:

  1. Understand how machine learning can be applied to polymer informatics.
  2. Implement regression models for predicting key polymer properties.
  3. Explore how high-throughput molecular dynamics data can assist in materials design.

📊 Dataset

Source

The dataset used in this project is derived from the PI1M database, which contains approximately 1 million polymer structures represented in p-SMILES format.

Format

  • File: PI1M_v2.csv
  • Columns: polymer p-SMILES, property values (density, Tg, Tm, dielectric constant), and SA score.

The synthetic accessibility score (SA) is computed using Schuffenhauer’s heuristic scoring approach
(Reference Paper).

🧩 Methodology

  • Data Preprocessing: Cleaning and encoding polymer structures.
  • Feature Extraction: Using polymer embedding representations (PE).
  • Model Training: Applying regression models (e.g., Random Forest, Gradient Boosting, Neural Networks).
  • Evaluation: Comparing predictions for accuracy and interpretability.

📈 Results

Our analysis reproduces and extends trends reported in the original PI1M study. The ML models successfully predict thermal and structural polymer properties, demonstrating the potential of data-driven polymer design.

🧪 Tools & Technologies

  • Python (NumPy, Pandas, Scikit-learn, Matplotlib)
  • Jupyter Notebooks
  • Molecular Dynamics data (from original research)
  • GitHub for version control

📚 Reference

If this dataset or research concept benefits your academic or research work, please cite the original publication:

Ruimin Ma, Tengfei Luo.
PI1M: A Benchmark Database for Polymer Informatics.
Journal of Chemical Information and Modeling (2021).
DOI: 10.1021/acs.jcim.0c00726

⚠️ Note

This repository is for academic and educational purposes only.
All original data and concepts belong to the respective authors of the PI1M database and publication.
This work extends their dataset for learning and demonstration under fair-use for research.

👨‍💻 Author

Vivek Kumar
B.Tech, NIT Warangal
Branch: Biotechnology RollNumber: 21BTB0A81

**Harsh Hotala B.tech, NIT Warangal Branch: Electrical Engineering RollNumber: 22EEB0A64

Project: Machine Learning-Assisted Materials Informatics (MLMI)
Year: 2025