Multitask Learning Reveals Shared Descriptors Governing Activity and Selectivity in Catalytic Nitrate Reduction.

Area of Science:

• Materials Science
• Chemical Engineering
• Environmental Science

Background:

• Catalytic nitrate reduction (CNR) is crucial for water treatment, but catalyst development faces challenges due to the activity-selectivity trade-off.
• Existing methods for catalyst design in CNR are largely empirical, limiting efficiency and increasing ammonium byproduct formation.

Purpose of the Study:

• To develop a multitask Gaussian process (MTGP) framework for characterizing and optimizing the activity-selectivity trade-off in CNR.
• To improve catalyst design by jointly modeling catalytic activity and ammonium selectivity using literature-curated data.

Main Methods:

• A multitask Gaussian process (MTGP) framework was developed to model catalytic activity and ammonium selectivity simultaneously.
• A large, literature-curated dataset of Palladium (Pd)- and Platinum (Pt)-based catalysts was utilized.
• A consensus feature-importance framework was employed to identify key catalytic descriptors for mechanistic interpretability.

Main Results:

• The MTGP framework successfully captured the coupling between catalyst properties and performance, improving predictive stability and generalization by 41-45% in ΔR² and up to 48% in ΔRMSE compared to single-task models.
• Key catalytic descriptors influencing N₂ versus NH₄⁺ formation were identified, aligning with established catalytic principles.
• External validation and knowledge transfer demonstrated the framework's predictive robustness and transferability.

Conclusions:

• Multitask learning provides a data-driven approach to capture transferable activity-selectivity relationships in CNR.
• The developed MTGP framework offers a powerful tool for guiding catalyst and process design in water treatment applications.
• This study enhances mechanistic understanding and predictive capabilities for developing more efficient nitrate removal catalysts.