First-principles, machine learning and symbolic regression modelling for organic molecule adsorption on...

Area of Science:

• Materials Science
• Computational Chemistry
• Data Science

Background:

• Traditional simulation methods are limited for modeling organic molecule adsorption on low-dimensional surfaces.
• Data-driven approaches offer a promising alternative for chemical and materials research.
• Understanding adsorption is crucial for designing new materials and predicting chemical interactions.

Purpose of the Study:

• To develop and apply data-driven methods for modeling organic molecule adsorption on low-dimensional metal oxide surfaces.
• To compare the performance of different machine learning algorithms for predicting adsorption energy.
• To identify key molecular descriptors and explore novel descriptors using symbolic regression.

Main Methods:

• Density Functional Theory (DFT) calculations to generate initial datasets of organic/metal oxide interfaces.
• Machine learning algorithms, including Random Forest, for predicting adsorption energy.
• Symbolic regression coupled with genetic programming for automated descriptor discovery.

Main Results:

• Random Forest algorithm demonstrated high accuracy in predicting adsorption energy.
• Polarizability and bond type of organic adsorbates were identified as key descriptors.
• Symbolic regression successfully generated new hybrid descriptors that improved prediction relevance.

Conclusions:

• A comprehensive data-driven framework was established for modeling and analyzing organic molecule adsorption on low-dimensional surfaces.
• Machine learning and symbolic regression can effectively complement traditional methods for materials discovery.
• The identified descriptors provide insights into the factors governing adsorption processes.