Tree based machine learning framework for predicting ground state energies of molecules

Area of Science:

• Computational chemistry
• Machine learning in chemistry

Background:

• Predicting molecular ground state energies is crucial for chemical research.
• Density functional theory (DFT) is computationally expensive for large datasets.

Purpose of the Study:

• To apply the boosted regression tree (BRT) algorithm for predicting molecular ground state energies.
• To assess the accuracy and computational efficiency of BRT compared to neural networks.

Main Methods:

• A dataset of 16,242 CHNOPS molecules was constructed from the PubChem database.
• Electronic ground state energies were computed using DFT.
• The BRT algorithm was trained on this dataset.

Main Results:

• BRT achieved accurate predictions of molecular ground state energies.
• BRT demonstrated superior accuracy and reduced computational cost compared to neural network regression.
• The model showed good performance on molecules containing Cl and Si.

Conclusions:

• BRT offers a computationally efficient and accurate approach for predicting molecular ground state energies.
• This method has broad applicability in molecular discovery and materials informatics.