Applying FAIR principles to materials and chemicals research data depends on rich annotation. Traditional schemas can block the ability to record the full context, whereas ontologies are cumbersome for the inexpert and still require substantial effort to develop mappings. A balance must therefore be struck between flexibility and standardization. When a domain expert builds a table for a particular application -- whether that be communication, critical evaluation or machine learning -- much data and metadata are omitted to conform to the chosen schema. It is vital that data storage and table building are separated to enable FAIR treatments.<br/><br/>Here, we discuss the implementation of the above concepts at Citrine Informatics. For our industrial partners, pursuit of FAIR principles is less about enabling interorganizational sharing and more about maximizing the return on investment for expensive experiments. Our approach allows partners to maintain full context for data <i>and</i> vectorize their data for various tasks. Our solution approach includes two layers: GEMD[1], a graph-based data structure where definitions of terminology are first class elements; and a data extraction layer defined in our citrine-python library [2]. First, GEMD presumes the history of any given sample can be represented as a directed acyclic graph, combining materials as ingredients to processes, which in turn yield new materials as a result. Attributes can be annotated on any object, with platform-defined templates serving the role of a controlled vocabulary, enforcing relationships between like concepts across very different processing histories. To enable Findability and Accessibility, objects can be annotated with UIDs and tags to support both unique and non-unique external identifiers. Second, the data extraction layer leverages the terminology, as defined in the templates, to serve as query parameters, much like the columns in a traditional SQL RDBMS. This separates aggregation and annotation from the structuring and normalization of table building, allowing the same base records to function as the digital underpinnings of multiple applications without constraining future annotations. Combined, these approaches enable forward and backward compatibility by storing data in a flexible format and applying repeatable simplifying transformations only later on.

Keep ‘em separated: GEMD data storage and their tabular offspring

ACS Spring 2022

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