Data-Driven Synthesis of Covalent Organic Frameworks via Machine Learning with Integrated Success-Failure Data

Area of Science:

• Materials Science
• Chemical Engineering
• Computational Chemistry

Background:

• Covalent organic frameworks (COFs) are porous materials with diverse applications.
• COF synthesis relies heavily on inefficient trial-and-error methods, hindering progress.
• A significant need exists for optimized and predictable COF synthesis strategies.

Purpose of the Study:

• To develop a machine learning (ML) model for predicting optimal COF synthesis conditions.
• To create a data-driven toolkit (ML-COF) to automate and improve COF synthesis.
• To provide a viable alternative to traditional, resource-intensive synthesis approaches.

Main Methods:

• Curated a dataset of 1822 failed and 2603 successful COF synthesis records.
• Employed a random forest ML model for its robustness with complex experimental data.
• Developed the ML-COF toolkit for automated route retrieval and condition prediction.

Main Results:

• Achieved 91% accuracy in solvent prediction for COF synthesis.
• Demonstrated high predictive performance for reaction temperature (R² = 0.9129) and time (R² = 0.9596).
• Experimental validation confirmed the toolkit's accuracy for eight challenging COFs.

Conclusions:

• A data-driven paradigm for COF synthesis has been established.
• The ML-COF toolkit significantly accelerates the discovery of novel COFs.
• This approach offers a practical alternative to trial-and-error synthesis, enabling future robot-assisted research.