Deep Generative Models for Molecular Science

Abstract

Generative deep machine learning models now rival traditional quantum-mechanical computations in predicting properties of new structures, and they come with a significantly lower computational cost, opening new avenues in computational molecular science. In the last few years, a variety of deep generative models have been proposed for modeling molecules, which differ in both their model structure and choice of input features. We review these recent advances within deep generative models for predicting molecular properties, with particular focus on models based on the probabilistic autoencoder (or variational autoencoder, VAE) approach in which the molecular structure is embedded in a latent vector space from which its properties can be predicted and its structure can be restored.

Inverse molecular design using machine learning: Generative models for matter engineering

“…Deep generative models are a special class of DL methods that seek to model the underlying probability distribution of both structure and property and relate them in a nonlinear way. By exploiting patterns in massive datasets, these models can distill average and salient features that characterize molecules (12,13).…”

ChemML: A Machine Learning and Informatics Program Package for the Analysis, Mining, and Modeling of Chemical and Materials Data

“…We will also advance the development of the overarching framework ChemEco that binds the different components of our software ecosystem [1] together and allows them to interact directly. Our long-term vision is to enable the fully automated exploration of compound space that supports the accelerated discovery and rational design of nextgeneration chemistry and materials [46,47].…”

Advances and challenges in deep generative models for de novo molecule generation

“…Basic AE, however, cannot be directly applied for de novo molecule generation because the model likely only learns some explicit mapping of the training data rather than a generalized sampling function of the molecule. Therefore, these models are modified with a constraint such as the variational autoencoder (VAE) or an adversarial autoencoder (AAE), to learn a latent variable z from the input data . The VAE provides a formulation in which the continuous representation z is interpreted as a latent variable in a probabilistic generative model.…”

“…De novo design of new molecules and analysis of their structure and properties is an important issue in computational molecular science. In the last few years, new approaches based on artificial intelligence (AI), especially deep learning models, have shown great promise for de novo molecular design and analysis . The deep learning model, which forms abstract representation learning on training samples (e.g., molecular representations) by cascading nonlinear feature transformations, enables efficient extraction of the underlying features of arbitrary input–output relationships, thereby facilitating quantitative structure–activity relationship analysis (QSAR) in computational molecular science .…”