Machine learning is increasingly being called upon to aid in new materials discovery.  Machine learning provide a means to map variations in experimental composition, reaction conditions, and physicochemical descriptors to possible reaction outcomes.  Building successful machine learning models requires the input data be structured such that humans can accurately input relevant information and computers can operate on the collected data.  Toward this end, we put forward the ESCALATE platform and associated ontology for building flexible data sets for machine learning.  The ESCALATE system has been applied to the development of two distinct materials research problems: high-throughput metal halide perovskite crystal discovery and the synthetic accessibility of polar racemates.  The first project, high-throughput data collection for perovskite discovery, required development of data-pipelines which integrated data storage methods, hardware-software interfaces, and simple experimental user interfaces. Since implementation, ESCALATE has enabled the discovery of 24 previously unreported perovskite systems with retrograde solubility, over 7,500 experimental runs, and real-time collaborative design-of-experiment campaigns with data scientists.  The success of the predictive models developed from the perovskite dataset have guided the synthesis of high quality single crystals in previously unexplored systems. In the second project, the ESCALTE ontology was used to construct and analyze polar racemate experiments.  These data enabled the development of robust visualizations, interpretable decision tree models, and unbiased assessments of human intuition through interpolative and extrapolative modeling (including gradient boosted trees, k-nearest neighbors, and support vector classifier models).  Perovskite and polar racemate materials discovery will be used to underscore the flexibility of the ESCALATE approach and enable a deeper discussion into the key aspects of organizing and managing experimental data at various scales.  This work provides an overview of the tools necessary for further development of real-time data capture in materials chemistry as well as outlining a path toward dataset construction and dissemination.

Preparing machine-learning-ready datasets using the ESCALATE ontology

Macromolecular Chemistry: The Second Century

[Author Benefits](https://scimeetings.acs.org/?utm_source=pubs_content_marketing&utm_medium=website&utm_campaign=0320_MCF_NPI_Launch_Spring_Homepage&ref=pubs_content_marketing)

[How to use SciMeetings](https://storage.googleapis.com/dl.morressier.com/acs/SciMeetings-ACSSpring2020-HowToGuide_M3020-Final.pdf)

[Frequently Asked Questions](https://storage.googleapis.com/dl.morressier.com/acs/SciMtng_ACSNMSp2020_FAQ_M3120_Final.pdf)

ACS Spring 2020 National Meeting & Expo

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.