AI-Driven Single-Atom Catalyst Discovery for CO₂ Reduction & Green Hydrogen: Machine Learning, Adsorption Energy & High-Throughput Screening

Aim

Workshop Objectives

Workshop Structure

🗓️ Day 1: Single-Atom Catalyst Representation & Machine Learning Activity Prediction

• Understanding the fundamentals of Single-Atom Catalysts (SACs), isolated active sites, metal–support interactions, and coordination environments such as M–N₄ and defect-engineered supports.
• Representing SAC structures, catalytic surfaces, active metal centres, and adsorbates as machine-readable materials descriptors.
• Computing coordination number, electronegativity, elemental, structural, and adsorption-related features for catalyst modelling.
• Building machine-learning surrogate models to predict adsorption and binding energies relevant to CO₂ reduction and hydrogen evolution.
• Evaluating model reliability using MAE, R², cross-validation, feature importance, and explainable-AI methods.

🧰 Tools

• Matminer: Materials descriptors and feature engineering.
• ASE: Atomic structure and catalyst-surface representation.
• Catalysis-Hub / Public Catalyst Datasets: Adsorption-energy and catalyst-property data.
• Scikit-learn / XGBoost: Machine-learning surrogate modelling.
• SHAP: Interpretation of catalyst descriptors and ML predictions.

🧪 Hands-on Lab (Google Colab)

• Notebook: Predicting Adsorption Energies of Single-Atom Catalysts for CO₂ Reduction Using Machine Learning.
• Task: Generate SAC descriptors, train an adsorption-energy prediction model, evaluate its accuracy, and identify the structural and chemical features governing catalytic activity.
• Deliverable: Validated SAC adsorption-energy model with MAE, R², cross-validation results, SHAP feature analysis, and an initial shortlist of promising single-atom catalysts.

🗓️ Day 2: Equivariant AI, Atomic Relaxation & Catalyst–Adsorbate Interactions

• Understanding graph-based and 3D equivariant representations of atomic structures for catalyst modelling.
• Representing SAC active sites, supports, and adsorbates as atomic graphs for energy and force prediction.
• Applying pretrained AI interatomic potentials to accelerate catalyst–adsorbate energy calculations and structural relaxation.
• Performing AI-guided geometry optimisation to identify stable catalyst–adsorbate configurations and local-energy minima.
• Comparing adsorption behaviour of key intermediates involved in CO₂RR, HER, and OER.
• Interpreting metal–adsorbate distances, coordination changes, predicted forces, and adsorption energies.

🧰 Tools

• FAIR-Chem / UMA: Pretrained universal models for atomic energy and force prediction.
• ASE: Construction, manipulation, and relaxation of catalyst–adsorbate structures.
• PyTorch / Graph-Based AI Concepts: Atomic graph representation and model interpretation.
• Py3Dmol / ASE Visualizer: Interactive visualization of relaxed SAC structures.

🧪 Hands-on Lab (Google Colab)

• Notebook: AI-Accelerated Atomic Relaxation of Single-Atom Electrocatalysts for CO₂RR and Green Hydrogen.
• Task: Construct a representative SAC–adsorbate system, perform AI-guided atomic relaxation, calculate adsorption energies, and compare competing catalytic intermediates.
• Deliverable: Optimised 3D SAC–adsorbate structures with predicted energies, forces, relaxation trajectories, and adsorption-energy comparison.

🗓️ Day 3: Virtual High-Throughput Screening, Selectivity & Multi-Objective Catalyst Ranking

• Designing Virtual High-Throughput Screening (VHTS) pipelines for large libraries of single-atom catalysts.
• Predicting and comparing activity descriptors for CO₂ Reduction Reaction (CO₂RR), Hydrogen Evolution Reaction (HER), and Oxygen Evolution Reaction (OER).
• Analysing competition between CO₂RR and HER to identify selective catalysts for carbon conversion.
• Generating adsorption-energy relationships, activity maps, and volcano-type plots for catalyst interpretation.
• Applying multi-objective optimisation using catalytic activity, selectivity, structural stability, metal cost, and elemental availability.
• Using explainability and uncertainty-aware filtering to distinguish robust catalyst candidates from unreliable ML predictions.
• Ranking SAC candidates and identifying high-priority structures for subsequent DFT calculation or experimental validation.

🧰 Tools

• FAIR-Chem / UMA: AI-based catalyst property and energy prediction.
• Catalysis-Hub: Catalyst and adsorption-energy datasets.
• XGBoost / Scikit-learn: Activity and selectivity prediction.
• SHAP: Explainable-AI analysis.
• Optuna: Multi-objective catalyst optimisation.
• Matplotlib / Pandas: Screening, ranking, and research-ready visualization.

🧪 Hands-on Lab (Google Colab)

• Notebook: Building an AI-Powered Virtual Screening Engine for Single-Atom Catalysts in CO₂ Reduction and Green Hydrogen.
• Task: Screen a virtual SAC library, predict adsorption/activity descriptors, analyse CO₂RR–HER selectivity, apply multi-objective filters, and rank candidates for further validation.
• Deliverable: Ranked shortlist of high-potential single-atom catalysts with activity/selectivity maps, explainable-AI results, optimisation scores, and publication-ready visualizations.

Who Should Enrol?

Important Dates

Workshop Outcomes