Machine Learning for Materials Science

A comprehensive, interactive learning path for applying machine learning to materials discovery, property prediction, and atomistic simulations.

Learning Path Overview

┌─────────────────────────────────────────────────────────────────────────────┐
│                    ML FOR MATERIALS SCIENCE                                 │
│                    Complete Learning Path                                   │
├─────────────────────────────────────────────────────────────────────────────┤
│                                                                             │
│  Tutorial 07          Tutorial 08              Tutorial 09                  │
│  ┌─────────────┐     ┌─────────────────┐      ┌──────────────────┐          │
│  │ ML          │     │ Neural Network  │      │ Advanced         │          │
│  │ Discovery   │ ──► │ Potentials      │ ──►  │ Features         │          │
│  │             │     │                 │      │                  │          │
│  │ • ML Basics │     │ • GNN Basics    │      │ • SOAP/MBTR      │          │
│  │ • matminer  │     │ • M3GNet/CHGNet │      │ • Active Learn   │          │
│  │ • sklearn   │     │ • MD with NNPs  │      │ • Bayesian Opt   │          │
│  │ • SHAP      │     │ • Fine-tuning   │      │ • Genertic Models│          │
│  └─────────────┘     └─────────────────┘      └──────────────────┘          │
│                                                                             │
│  Difficulty:  🟢 Beginner  →  🟡 Intermediate  →  🔴 Advanced                │
│                                                                             │
└─────────────────────────────────────────────────────────────────────────────┘

Prerequisites

Before starting, you should be comfortable with:

• Python basics: variables, functions, loops, classes
• NumPy & Pandas: array operations, DataFrames
• Basic chemistry/materials science: what are crystals, compositions, properties
• Optional but helpful: Previous tutorials (01-06) in this series

Quick Start

Option 1: Cloud (No Installation Required)

Each notebook includes buttons to launch in:

• Google Colab: Click "Open in Colab" badge
• Binder: Click "Launch Binder" badge

Option 2: Local Installation

# Clone the repository
git clone https://github.com/NabKh/ML-for-Materials-Science.git
cd ML-for-Materials-Science

# Create conda environment
conda env create -f environment.yml
conda activate ml-materials

# Verify installation
jupyter notebook setup_check.ipynb

# Start learning!
jupyter lab

GPU Support (Optional)

For faster neural network training:

# After activating environment
conda install pytorch pytorch-cuda=12.1 -c pytorch -c nvidia

Tutorial Structure

Tutorial 07: ML for Materials Discovery

Foundation course - Start here!

#	Notebook	Difficulty	Key Topics
1	ML Fundamentals	🟢	Supervised/unsupervised, overfitting, cross-validation
2	Data Foundation	🟢	Materials Project API, data cleaning, splits
3	Featurization Basics	🟢🟡	matminer, composition & structure features
4	Classical ML Models	🟡	Linear → Random Forest → XGBoost
5	Model Evaluation	🟡	Metrics, learning curves, hyperparameter tuning
6	Explainable AI	🟡🔴	SHAP values, feature importance
7	Project: Band Gap	🔴	End-to-end ML pipeline

Tutorial 08: Neural Network Potentials

Deep learning for atomistic simulations

#	Notebook	Difficulty	Key Topics
1	Why NNPs?	🟢	DFT limitations, accuracy vs speed
2	GNN Basics	🟡	Graphs, message passing, CGCNN
3	Universal MLIPs	🟡	M3GNet, CHGNet, MACE architectures
4	Pretrained Models	🟡	MatGL, loading models, predictions
5	MD with NNPs	🟡🔴	ASE integration, simulations
6	Fine-tuning	🔴	Transfer learning, avoiding forgetting
7	Project: Phonons	🔴	Phonon calculation with NNPs

Tutorial 09: Advanced Features & Discovery

Cutting-edge ML for materials

#	Notebook	Difficulty	Key Topics
1	Atomic Descriptors	🟡	SOAP, MBTR, ACSF with DScribe
2	Electronic Features	🟡	DOS fingerprints, band structure
3	Dimensionality Reduction	🟡	PCA, t-SNE, UMAP visualization
4	Active Learning	🟡🔴	Bayesian optimization, acquisition
5	Multi-objective Opt	🔴	Pareto fronts, trade-offs
6	Generative Models	🔴	VAE, diffusion intro
7	Project: Alloy Design	🔴	Design alloy with target properties

Interactive Features

Jupyter Widgets

Every notebook includes interactive elements:

• Sliders to explore hyperparameters
• Dropdowns to select models/features
• Checkboxes for feature selection
• Interactive plots with Plotly

Self-Check Quizzes

Test your understanding with embedded quizzes:

# Example quiz widget
quiz.check_answer("What prevents overfitting?", your_answer)

Visual Learning

• Animated diagrams explaining concepts
• Side-by-side model comparisons
• Interactive feature importance plots
• 3D materials space exploration

Key Libraries Used

Library	Purpose	Documentation
scikit-learn	Classical ML algorithms	docs
matminer	Materials featurization	docs
pymatgen	Materials analysis	docs
dscribe	Atomic descriptors	docs
matgl	Graph neural networks	docs
shap	Model explainability	docs
mp-api	Materials Project API	docs

Data Sources

Database	Content	Access
Materials Project	~150,000 materials	API key (free)
AFLOW	~3.5M materials	Open
OQMD	~1M materials	Open
JARVIS-DFT	~75,000 materials	Open

Note: You'll need a free Materials Project API key. Get one at: https://materialsproject.org/api

Contributing

Found an error? Have a suggestion? Please open an issue or pull request!

Citation

If you use these tutorials in your research or teaching, please cite:

Khossossi, N. (2026). ML for Materials Science: Interactive Tutorial Series.
https://sustai-nabil.com/teaching

License

This work is licensed under CC BY-NC-SA 4.0. You are free to share and adapt for non-commercial purposes with attribution.

Acknowledgments

• Materials Project team for pymatgen and matminer
• MatGL developers for M3GNet/CHGNet implementations
• DScribe team for atomic descriptors
• The open-source ML and materials science communities

---

Ready to start? Open Tutorial-07-ML-Discovery/notebooks/01_ml_fundamentals.ipynb

Questions? Contact: n.khossossi@differ.nl