ABSTRACT
The ever-growing abundance of data found in heterogeneous sources, such as scientific publications, has forced the development of automated techniques for data extraction. While in the past, in the physical sciences domain, the focus has been on the precise extraction of individual properties, attention has recently been devoted to the extraction of higher-level relationships. Here, we present a framework for an automated population of ontologies. That is, the direct extraction of a larger group of properties linked by a semantic network. We exploit data-rich sources, such as tables within documents, and present a new model concept that enables data extraction for chemical and physical properties with the ability to organize hierarchical data as nested information. Combining these capabilities with automatically generated parsers for data extraction and forward-looking interdependency resolution, we illustrate the power of our approach via the automatic extraction of a crystallographic hierarchy of information. This includes 18 interrelated submodels of nested data, extracted from an evaluation set of scientific articles, yielding an overall precision of 92.2%, across 26 different journals. Our method and associated toolkit, ChemDataExtractor 2.0, offers a key step toward the seamless integration of primary literature sources into a data-driven scientific framework.
Subject(s)
Materials Science , Software , Information Storage and RetrievalABSTRACT
Generative models have been successfully used to synthesize completely novel images, text, music, and speech. As such, they present an exciting opportunity for the design of new materials for functional applications. So far, generative deep-learning methods applied to molecular and drug discovery have yet to produce stable and novel 3-D crystal structures across multiple material classes. To that end, we, herein, present an autoencoder-based generative deep-representation learning pipeline for geometrically optimized 3-D crystal structures that simultaneously predicts the values of eight target properties. The system is highly general, as demonstrated through creation of novel materials from three separate material classes: binary alloys, ternary perovskites, and Heusler compounds. Comparison of these generated structures to those optimized via electronic-structure calculations shows that our generated materials are valid and geometrically optimized.
Subject(s)
Deep Learning , Drug Discovery , LearningABSTRACT
Large auto-generated databases of magnetic materials properties have the potential for great utility in materials science research. This article presents an auto-generated database of 39,822 records containing chemical compounds and their associated Curie and Néel magnetic phase transition temperatures. The database was produced using natural language processing and semi-supervised quaternary relationship extraction, applied to a corpus of 68,078 chemistry and physics articles. Evaluation of the database shows an estimated overall precision of 73%. Therein, records processed with the text-mining toolkit, ChemDataExtractor, were assisted by a modified Snowball algorithm, whose original binary relationship extraction capabilities were extended to quaternary relationship extraction. Consequently, its machine learning component can now train with ≤ 500 seeds, rather than the 4,000 originally used. Data processed with the modified Snowball algorithm affords 82% precision. Database records are available in MongoDB, CSV and JSON formats which can easily be read using Python, R, Java and MatLab. This makes the database easy to query for tackling big-data materials science initiatives and provides a basis for magnetic materials discovery.
