Solid harmonic wavelet scattering for predictions of molecule properties.

Area of Science:

• Computational chemistry
• Machine learning
• Quantum mechanics

Background:

• Predicting molecular properties is crucial for drug discovery and materials science.
• Traditional methods like density functional theory (DFT) are computationally expensive.
• Machine learning offers a potential avenue for faster property prediction.

Purpose of the Study:

• To develop a novel machine learning algorithm for predicting molecular properties.
• To leverage concepts from DFT for improved accuracy and interpretability.
• To create a computationally efficient method for molecular property prediction.

Main Methods:

• Utilized Gaussian-type orbital functions to generate surrogate electronic densities.
• Computed invariant "solid harmonic scattering coefficients" capturing multi-scale interactions.
• Employed multilinear regressions on these coefficients to predict physical properties.

Main Results:

• The developed algorithm demonstrates near state-of-the-art performance.
• Achieved high accuracy even with limited training data.
• Predictions using small sets of coefficients reached DFT precision and remained interpretable.

Conclusions:

• The machine learning algorithm effectively predicts molecular properties.
• The method offers a balance of accuracy, interpretability, and computational efficiency.
• This approach shows promise for accelerating molecular property prediction in computational chemistry.