Vast information on existing crystal structures, which is available through the large open-access and commercialized repositories of crystallographic data, provides an excellent starting point for the implementation of deep learning techniques to the discovery of hidden relationships that might be contained in such large datasets. The machine learning algorithm can be thought of as a fitting procedure for a complicated heuristic model using a large amount of data.1-2 This model is later tested to estimate its ability to generalize to unknown crystal structures in a holdout set, i.e. its predictive ability. Herein, we describe a neural network model trained to predict the likelihood of chemical elements adopting different topologies of atomic sites in known crystal structures. The neural network is shown examples of topologies from known crystal structures and trained to predict the element that adopted that topology. We apply the trained model to predict possible compositions of unknown compounds that might be pursued by a synthetic chemist. We demonstrate that the deep neural network is capable of automatically “discovering” relevant descriptors from high-dimensional “raw representations” of the crystallographic data. Since the input data contain purely geometrical and topological information, any chemical knowledge residing within the neural network output must have been learned during training, and thus was “discovered”. The neural network’s learned representation of local topology shows evidence of known geometric and chemical trends not explicitly provided to the network during training. / A Dissertation submitted to the Department of Chemistry and Biochemistry in partial fulfillment of the requirements for the degree of Doctor of Philosophy. / Summer Semester 2018. / July 10, 2018. / Crystallography, Deep Learning / Includes bibliographical references. / Mykhailo Shatruk, Professor Directing Dissertation; Adrian Gheorghe Barbu, University Representative; Albert E. De Prince, Committee Member; Susan E. Latturner, Committee Member.
| Identifer | oai:union.ndltd.org:fsu.edu/oai:fsu.digital.flvc.org:fsu_647293 |
| Contributors | Ryan, Kevin (author), Shatruk, Mykhailo (professor directing dissertation), Barbu, Adrian G., 1971- (university representative), DePrince, A. Eugene (committee member), Latturner, Susan (committee member), Florida State University (degree granting institution), College of Arts and Sciences (degree granting college), Department of Chemistry and Biochemistry (degree granting departmentdgg) |
| Publisher | Florida State University |
| Source Sets | Florida State University |
| Language | English, English |
| Detected Language | English |
| Type | Text, text, doctoral thesis |
| Format | 1 online resource (80 pages), computer, application/pdf |
Page generated in 0.0016 seconds