Deep Learning and Extraction of Chemical Synthesis or Biosynthetic Pathways from Scientific Literature

This SBIR Phase I project proposes novel deep learning models and training techniques for the extraction of chemical/biochemical synthesis pathways from written works. The proposed system allows for the agnostic textual mining of reaction data and its integration with a functional database. The automation of such a pipeline allows for the live monitoring of potential threats and prediction of novel synthesis pathways from extracted data. Monitoring of scientific literature including journal articles and patents will allow for the identification of dual-use research of concern (DURC) where cutting-edge research may have both beneficial and harmful applications. The Phase I effort will focus on the development of textual mining of chemical synthesis data for literary sources and its integration into a database for identification of concerning trends and threats within scientific literature. The team will benefit from its ample experience in the areas of deep learning networks, chemistry, database engineering, and large-language model text mining.