Graph neural networks for materials science and chemistry

Abstract

Machine learning plays an increasingly important role in many areas of chemistry and materials science, being used to predict materials properties, accelerate simulations, design new structures, and predict synthesis routes of new materials. Graph neural networks (GNNs) are one of the fastest growing classes of machine learning models. They are of particular relevance for chemistry and materials science, as they directly work on a graph or structural representation of molecules and materials and therefore have full access to all relevant information required to characterize materials. In this Review, we provide an overview of the basic principles of GNNs, widely used datasets, and state-of-the-art architectures, followed by a discussion of a wide range of recent applications of GNNs in chemistry and materials science, and concluding with a road-map for the further development and application of GNNs.

Machine learning-assisted materials development and device management in batteries and supercapacitors: performance comparison and challenges

“…Future research should focus on the continued development of novel architectures that incorporate physical principles and additional material information, as well as the development of a labeled materials dataset for model training. 194…”

“…Future research should focus on the continued development of novel architectures that incorporate physical principles and additional material information, as well as the development of a labeled materials dataset for model training. 194 NLP capabilities for data extraction need improvement: NLP text mining has promising applications for energy materials research, especially for creating more comprehensive training datasets that could improve ML model accuracy. NLP greatly depends on the quality of the input text, which means data labels and the amount of data that can be extracted from articles depending on the details provided and terms used by the authors.…”

Representations of Materials for Machine Learning

“…In the previous section, we described many physically-inspired descriptors that characterize materials and can be used to efficiently predict properties. The use of differentiable graph-based representations in convolutional neural networks, however, mitigates the need for manual engineering of descriptors [53,54]. Indeed, advances in deep learning and the construction of large-scale materials databases [4,5,6,7,8,9] have made it possible to learn representations directly from structural data.…”

“…Greater knowledge of local distortions introduced at varying grain boundary incident angles would give computational materials scientists a more complete understanding of how experimentally chosen chemistries and synthesis parameters will translate into device performance. Strategies to quantify characteristics of grain boundary geometry have included reducing computational requirements by identifying the most promising configurations with virtual screening [117], estimating grain boundary free volume as a function of temperature and bulk composition [118], treating the microstructure as a graph of nodes connected across grain boundaries [119,54], and predicting the energetics, and hence feasiblity, of solute segregation [120]. While the previous approaches did not include features based on the constituent atoms and were only benchmarked on systems with up to three elements, recent work has demonstrated that the excess energy of the grain boundary relative to the bulk can be approximated across compositions with five variables defining its orientation and the bond lengths within the grain boundary (Figure 4c) [121].…”

High-throughput synthesis and machine learning assisted design of photodegradable hydrogels

“…by improving the virtual design of materials [20] , speeding up optimization processes [21][22][23] or yielding a better understanding of hidden relations in data and thus fundamental processes. [24][25][26][27] With continuing progress in ML research, the research speed of manual lab experiments and thus the generation of new data is not sufficient any longer to sustain the amounts of data needed to efficiently make use of new ML methods. Thus, increasing the throughput of experimental pipelines comes as a logical consequence and drives the research community to parallelize experiments in order to speed up data generation.…”