Aims and Fit of Module

This module covers the basic mechanisms of advanced AI algorithms (graph neural networks, transformer, etc.) and as well as hands-on case studies to empower students to harness powerful open-source AI tools to accelerate materials discovery, predict properties with unprecedented accuracy, and mine scientific literature at scale. Through a combination of interactive coding labs, case studies from top open AI tools, and a capstone group project, students will develop the skills to become pioneers in this transformative intersection of disciplines.
Tool Selection Rationale: This course leverages at least three specialised open-source tools—MATGL for crystal property prediction using graph networks, DeepChem for molecular machine learning, and large language model packages for materials text mining—because they represent the gold standard in AI-driven materials research. These tools were developed by leading institutions (Berkeley Lab, MIT, and LLNL), and these frameworks combine cutting-edge algorithms with domain-specific optimisations, enabling accurate modelling of materials from atomic structures to published literature. Their Python-based ecosystems ensure seamless integration with modern ML workflows while maintaining accessibility for advanced study. By adopting these tools, students gain hands-on experience with the same technologies used in groundbreaking research and industry applications, preparing them to contribute to the next wave of materials innovation.

Learning outcomes

A. Describe the principles of AI algorithms (GNNs, transformers) and their applications in materials science
B. Implement graph neural networks for crystal structure property prediction.
C. To build models for molecular property prediction (e.g., solubility, toxicity).
D. Utilise large language models to extract materials-related information from scientific literature.
E. Design and execute a group project to solve a materials science problem using AI tools.
F. Apply the teamwork, time management, presentation and reporting skills.

Method of teaching and learning

• Lectures: Interactive sessions with live demos (e.g., Google Colab; Jupyter notebooks for implementing open-source AI tools designed specifically for materials science).
• Tutorials and Labs: Code reviews, debugging, and case study discussions.
• Group Design Projects: Peer-assessed, with milestones (proposal, interim, final pitch) in materials science and engineering.