[Paper Review] Convolutional Neural Networks and Volcano Plots: Screening and Prediction of Two-Dimensional Single-Atom Catalysts

Haoyu Yang, Juanli Zhao+5 more|arXiv (Cornell University)|Feb 6, 2024

Machine Learning in Materials ScienceMaterials Science3 citations

TL;DR

This study proposes a hybrid framework integrating multi-branch convolutional neural networks (CNNs) with volcano plots to screen and predict two-dimensional single-atom catalysts (SACs) for CO2 reduction. By using electronic density of states (eDOS) as input, the CNN predicts adsorption energies with a mean absolute error of 0.1 eV, enabling accurate identification of optimal SACs for CH4 production via orbital-wise occlusion and shifting experiments.

ABSTRACT

Single-atom catalysts (SACs) have emerged as frontiers for catalyzing chemical reactions, yet the diverse combinations of active elements and support materials, the nature of coordination environments, elude traditional methodologies in searching optimal SAC systems with superior catalytic performance. Herein, by integrating multi-branch Convolutional Neural Network (CNN) analysis models to hybrid descriptor based activity volcano plot, 2D SAC system composed of diverse metallic single atoms anchored on six type of 2D supports, including graphitic carbon nitride, nitrogen-doped graphene, graphene with dual-vacancy, black phosphorous, boron nitride, and C2N, are screened for efficient CO2RR. Starting from establishing a correlation map between the adsorption energies of intermediates and diverse electronic and elementary descriptors, sole singular descriptor lost magic to predict catalytic activity. Deep learning method utilizing multi-branch CNN model therefore was employed, using 2D electronic density of states as input to predict adsorption energies. Hybrid-descriptor enveloping both C- and O-types of CO2RR intermediates was introduced to construct volcano plots and limiting potential periodic table, aiming for intuitive screening of catalyst candidates for efficient CO2 reduction to CH4. The eDOS occlusion experiments were performed to unravel individual orbital contribution to adsorption energy. To explore the electronic scale principle governing practical engineering catalytic CO2RR activity, orbitalwise eDOS shifting experiments based on CNN model were employed. The study involves examining the adsorption energy and, consequently, catalytic activities while varying supported single atoms. This work offers a tangible framework to inform both theoretical screening and experimental synthesis, thereby paving the way for systematically designing efficient SACs.

Motivation & Objective

• To overcome the limitations of traditional descriptor-based methods in screening single-atom catalysts (SACs) for CO2 reduction.
• To address the high computational cost and complexity of density functional theory (DFT) calculations in evaluating SAC performance.
• To develop a data-driven, automated method for identifying optimal SACs with high activity and selectivity toward CH4 production.
• To uncover the electronic structure contributions to adsorption energy through explainable AI techniques like eDOS occlusion and shifting.
• To establish a systematic framework for guiding both theoretical screening and experimental synthesis of efficient 2D SACs.

Proposed method

• Employed a multi-branch CNN model that takes 2D electronic density of states (eDOS) as input to predict adsorption energies of CO2RR intermediates.
• Integrated the CNN predictions into a hybrid descriptor-based volcano plot that includes both C- and O-type intermediates for improved activity screening.
• Conducted eDOS occlusion experiments with varying mask widths (11–51) to identify orbital contributions to adsorption energy.
• Performed orbitalwise eDOS shifting experiments to probe the sensitivity of adsorption energy to energy-level shifts, using a 0.005 eV resolution.
• Used HyperBand hyperparameter optimization via KerasTuner on NVIDIA V100 GPUs to train the CNN model with high precision.
• Validated model performance against DFT-calculated adsorption energies, achieving a mean absolute error (MAE) of 0.1 eV.

Figure 1: Catalyst performance analysis pipeline integrating volcano plots and CNN. The figure demonstrates the integrated usage of volcano plots for predictive assessment of existing catalysts, and the utility of CNNs to predict and modulate adsorption energies $E_{\text{ads}}$ using eDOS as input.

Experimental results

Research questions

• RQ1 Can a deep learning model trained on eDOS accurately predict adsorption energies of CO2RR intermediates in 2D single-atom catalysts?
• RQ2 How do individual electronic orbitals contribute to the adsorption energy of key reaction intermediates in SACs?
• RQ3 To what extent does shifting the energy level of specific orbitals affect the predicted catalytic activity of SACs?
• RQ4 Can a hybrid volcano plot incorporating both C- and O-type intermediates improve the screening of efficient CH4-producing catalysts?
• RQ5 Can explainable AI techniques like eDOS occlusion and shifting reveal the underlying electronic structure principles governing SAC performance?

Key findings

• The CNN model achieved a mean absolute error (MAE) of 0.1 eV in predicting adsorption energies for nine CO2RR and HER intermediates, matching the accuracy of conventional DFT calculations.
• eDOS occlusion experiments revealed that the d-orbitals of the central metal atom and p-orbitals of the support material significantly influence adsorption energy, especially near the Fermi level.
• Orbitalwise eDOS shifting experiments showed that shifting the d-band center by 0.005 eV alters the predicted adsorption energy by up to 0.15 eV, highlighting the sensitivity of catalytic activity to electronic structure tuning.
• The hybrid volcano plot successfully identified promising SAC candidates—such as Ge@g-C3N4 and Sn@BN—for CH4 production with low overpotential and high selectivity.
• The framework enables systematic exploration of electronic structure-activity relationships, providing actionable insights for catalyst design beyond empirical descriptor methods.
• The study demonstrates that deep learning models trained on eDOS can autonomously extract complex structure-activity relationships without manual descriptor engineering.

Figure 2: CNN for electronic density of states. a . Kendall rank correlation coefficient map illustrating the correlation between adsorption energy and other electronic and elementary descriptors. Definitions of the notations are provided in Supplementary Table 15 . b - c . Architectural of ( b ) th