DTRA DTRA234-003 Deep Learning 23.4

Knowledge Base Systems, Inc. (KBSI) proposes design and development of a deep learning-based Predictive Analytics Technology for Discovering Enhanced Chemical and Biosynthetic Routes (PATHFINDER).  PATHFINDER will harness the power of large language models (LLM) in conjunction with specialized expert tools to parse scientific literature and patents to accurately extract synthesis pathways.  By combining chain-of-thought reasoning with task specific tools, this capability will transition to accurate automated reasoning. PATHFINDER will integrate knowledge extracted from publications and patents and synthesis knowledge bases in a generative-based information extraction process so as to improve the accuracy of chemical and biosynthetic pathways by efficiently compensating for the deficiencies of LLMs in the specific using prompt chaining. Phase I of the proposed effort will focus on chemical synthesis knowledge extraction and capture in a scalable ontology.  Phase II will extend the scope to biosynthesis and identification of DURC.  The knowledge base and ontology extracted from scientific literature will be supplemented by open-source synthesis data sources.  The performance of the knowledge extraction will be verified and reported using precision and recall metrics against subject matter expert (SME) -validated manually tagged corpus.  Performance will also be compared with open source solutions.