An efficient catalyst screening strategy combining machine learning and causal inference.

Area of Science:

• Catalysis
• Materials Science
• Chemical Engineering

Background:

• Traditional catalyst optimization methods face challenges due to diverse synthesis routes.
• Efficient catalyst screening is crucial for developing advanced materials and processes.

Purpose of the Study:

• To develop a new strategy for determining catalyst performance by integrating causal inference with machine learning.
• To explore the relationship between nitrogen functional groups in catalysts and degradation performance for efficient catalyst screening.

Main Methods:

• A dataset of 14 parameters from 182 experiments was compiled, including catalyst properties and reaction conditions.
• Machine learning models, particularly CatBoost, were employed to predict catalyst performance.
• SHAP and DoWhy causal inference were used to identify key functional groups and their causal effects.

Main Results:

• The CatBoost model achieved high accuracy (R² = 0.953, MAE = 3.277, RMSE = 5.615) in predicting catalyst performance.
• SHAP analysis identified pyridinic N as a key nitrogen functional group influencing Bisphenol A (BPA) degradation.
• Causal inference confirmed the positive impact of pyridinic N on degradation performance, with an estimated causal effect of 0.4388.

Conclusions:

• The proposed strategy significantly enhances catalyst selection efficiency by reducing the process from multiple steps to one.
• Integrating causal inference with machine learning provides a powerful approach for optimizing catalysts and accelerating materials discovery.