About Workshop
This advanced three-day workshop introduces AI-guided catalyst discovery for sustainable industrial chemical processes. Participants will learn to represent catalyst structures, predict adsorption energies, perform AI-assisted atomic relaxation and build virtual high-throughput screening pipelines.
Hands-on Google Colab exercises will cover applications in CO₂ reduction, green hydrogen and electrocatalysis using tools such as Matminer, DeepChem, ASE, FAIR-Chem, PyTorch Geometric and Hugging Face Transformers.
Aim
To provide practical knowledge of integrating artificial intelligence, materials informatics and computational catalysis for faster catalyst screening and candidate prioritisation.
What Participants Will Learn
Participants will learn to:
- Convert catalyst structures and adsorbates into machine-readable descriptors.
- Build ML models for adsorption-energy prediction.
- Apply graph neural networks and pretrained interatomic potentials.
- Perform AI-guided catalyst–adsorbate structure relaxation.
- Develop multi-objective virtual screening workflows.
- Rank catalysts by activity, selectivity, stability, cost and material availability.
Structure
🗓️ Day 1: Material Representation & Machine Learning Surrogate Models
- Understanding the limitations of traditional Density Functional Theory (DFT) for large-scale catalyst discovery.
- Vectorizing solid-state crystal structures, adsorbates, and catalytic surface facets for machine learning models.
- Computing electronic descriptors, coordination numbers, and site-specific electronegativity features.
- Building, training, and benchmarking machine learning surrogate models to predict catalyst binding energies.
- Evaluating model reliability using Mean Absolute Error (MAE), R² scores, and cross-validation metrics.
🧰 Tools
- Matminer: Materials descriptors and feature engineering.
- RDKit: Molecular representation and chemical descriptor generation.
- DeepChem: Machine learning model development for chemical and materials data.
🧪 Hands-on Lab (Google Colab)
- Notebook: Predicting CO₂ Reduction Binding Energy Using Matminer Descriptors and DeepChem ML.
- Task: Generate catalyst descriptors and train a surrogate model to predict adsorption and binding energies.
- Deliverable: Produce a validated catalyst-property prediction model with MAE, R², and cross-validation results.
🗓️ Day 2: 3D Graph Neural Networks & AI-Guided Structure Relaxation
- Understanding 3D rotation equivariance and graph-based representations of atomic coordinates.
- Applying Graph Neural Networks (GNNs) to model complex atomic interactions and potential-energy surfaces.
- Replacing computationally expensive Quantum ESPRESSO and VASP calculations with neural network force fields.
- Running automated 3D surface–adsorbate structural relaxations to identify local energy minima.
- Accelerating energy-relaxation calculations by up to 1,000× for alloy and metal-oxide catalyst interfaces.
🧰 Tools
- Open Catalyst Project: Pre-trained checkpoints for catalyst energy and force prediction.
- ASE: Atomic structure construction, manipulation, simulation, and relaxation.
- PyTorch Geometric: Graph neural network development for atomic systems.
- EquiformerV2: Equivariant transformer architecture for accurate 3D atomic modelling.
🧪 Hands-on Lab (Google Colab)
- Notebook: 1,000× Accelerated Atomic Relaxation for Green H₂ Electrocatalysts Using EquiformerV2.
- Task: Construct a catalyst–adsorbate system and perform AI-guided atomic structure relaxation.
- Deliverable: Generate an optimized 3D catalyst structure with predicted energies, forces, and relaxation trajectories.
🗓️ Day 3: Generative AI, Chemical Transformers & Virtual High-Throughput Screening
- Understanding language models trained on SMILES, SELFIES, and materials representations.
- Fine-tuning pre-trained chemical transformer architectures for catalytic activity and property prediction.
- Designing multi-objective Virtual High-Throughput Screening (VHTS) pipelines for catalyst discovery.
- Filtering candidate libraries based on predicted overpotential, selectivity, binding energy, and structural stability.
- Ranking computational catalyst candidates using performance, reliability, and feasibility criteria.
- Formatting computational discovery workflows and visualizations for high-impact research publications.
🧰 Tools
- Hugging Face Transformers: Access to ChemBERTa, MolFormer, and related chemical language models.
- PyXSF / ASE Visualizer: Visualization of atomic structures, surfaces, and catalyst candidates.
- Open Catalyst API: Access to catalyst structures, energies, and pre-trained model predictions.
🧪 Hands-on Lab (Google Colab)
- Notebook: Fine-Tuning ChemBERTa and Building a Virtual High-Throughput Catalyst Screening Engine.
- Task: Fine-tune a chemical transformer and screen a virtual catalyst library against multiple performance criteria.
- Deliverable: Generate a ranked shortlist of computational catalyst candidates with predicted properties and publication-ready visualizations.
Important Dates
Registration Ends
7 September 2026 4:30 PM
Workshop Dates
7 September 2026 5:30 PM
What You Will Gain
Outcomes
By the end of the workshop, participants will be able to:
- Prepare catalyst datasets for machine learning.
- Train and evaluate catalyst-property prediction models.
- Use AI models for atomic structure and energy prediction.
- Screen large catalyst libraries computationally.
- Identify promising candidates for further DFT or experimental validation.
- Generate publication-ready catalyst-ranking reports and visualisations.
Who Should Attend
This workshop is suitable for postgraduate students, PhD scholars, faculty members, researchers and industry professionals working in:
- Chemical engineering
- Computational chemistry
- Materials science
- Heterogeneous catalysis
- Electrocatalysis
- Surface science
- Green hydrogen
- CO₂ conversion
- Nanocatalysis
- Materials informatics
- AI-driven materials discovery
- Sustainable industrial processes