Hierarchical modeling of molecular energies using a deep neural network

TL;DR

The paper presents HIP-NN, a hierarchical neural network model inspired by many-body expansion, achieving state-of-the-art accuracy in predicting molecular energies and aiding in uncertainty estimation.

Contribution

Introduces HIP-NN, a novel hierarchical neural network architecture for molecular energy prediction, inspired by many-body expansion, with improved accuracy and uncertainty detection.

Findings

Achieves 0.26 kcal/mol mean absolute error on organic molecules dataset.

Performs competitively on molecular dynamics datasets.

Hierarchical structure aids in identifying model uncertainty.

Abstract

We introduce the Hierarchically Interacting Particle Neural Network (HIP-NN) to model molecular properties from datasets of quantum calculations. Inspired by a many-body expansion, HIP-NN decomposes properties, such as energy, as a sum over hierarchical terms. These terms are generated from a neural network--a composition of many nonlinear transformations--acting on a representation of the molecule. HIP-NN achieves state-of-the-art performance on a dataset of 131k ground state organic molecules, and predicts energies with 0.26 kcal/mol mean absolute error. With minimal tuning, our model is also competitive on a dataset of molecular dynamics trajectories. In addition to enabling accurate energy predictions, the hierarchical structure of HIP-NN helps to identify regions of model uncertainty.