Semantically Structured Knowledge Graphs for the Inverse Design of Molecules

Natural Language Processing (NLP), which broadly encompasses language modeling techniques including large language models, automates the construction of machine-readable representations of text discourse. A Knowledge Graph (KG) is a graph representation of entities and their relationships. In this talk, we describe an approach that translates these methods to work together to process domain-specific text, resulting in a novel capability to create Semantically Structured Knowledge Graphs (SSKG) directly from highly technical language. The work is suited for situations where the numbers of documents are too many for a single individual to process. We then demonstrate how the resulting knowledge graph enables an unconventional inverse design where the knowledge in text is incorporated into the design process. The talk describes the approach using an exemplar for the inverse design of organic molecules. Through a set of queries that are uniquely enabled by the SSKG, we show how the design of molecules can be informed by the information in vast numbers of text sources.