SBIR-STTR Award

Knowledge Based Systems, Inc. (KBSI) proposes to design and develop Semantic Analysis Technologies for the Identification of Dual Use Research of Concern (STIR). The proposed STIR will process scientific documents using semantic technologies and inference algorithms to identify potential for Dual Use Research of Concern (DURC). The focus will be on 15 high consequence pathogens and toxins and seven experimental categories identified as DURC. STIR evaluates multiple semantic processing and inference approaches to generate the optimal DURC identification based on adaptability, precision, recall, time and effort required of subject matter experts, and ease of use. The methodology will be configured for each of the seven DURC experimental categoriesdue to variation in the information content and DURC identification requirements. Deep semantics (extracting relationships at the molecular and cellular level, identifying biochemical entities, species, hosts) and shallow semantics (looking for presence of identified pathogens and concepts associated with different experimental conditions within a sentence, neighborhood of a sentence, paragraph, or a document) will be explored for DURC identification and their performance will be analyzed. Fusion of the results of different DURC identification models will be performed to optimize the overall DURC identification.

The goal of this project is to design and develop Semantic Analysis Technologies for the Identification of Dual Use Research of Concern (STIR). STIR processes scientific documents using semantic technologies and inference algorithms to identify potential for Dual Use Research of Concern (DURC). The focus is on 15 high consequence pathogens and toxins and 7 experimental categories identified as DURC. STIR evaluates multiple semantic processing and inference approaches to generate the optimal DURC identification based on adaptability, precision, recall, time and effort required of subject matter experts, and ease of use. The methodology is configured for each of the seven DURC experimental categories – due to variation in the information content and DURC identification requirements. Deep semantics (extracting relationships at the molecular and cellular level, identifying biochemical entities, species, hosts, and their relationships) and shallow semantics (looking for presence of identified pathogens, hosts, and DURC experimental concepts within a sentence, neighborhood of a sentence, paragraph, or a document) are explored for DURC identification and their performance is analyzed. Fusion of the results of different DURC identification models is performed to optimize the overall DURC identification.