SBIR-STTR Award

The dual use potential of certain life sciences research has been recognized as an important biosecurity issue for a number of years. Current Dual Use Research of Concern (DURC) assessment procedures are pure manual work done by researchers/reviewers trained in the biosecurity domain. Therefore, an automated approach to analyze and identify DURC rapidly and reliably is urgently needed. To address this need, Intelligent Automation, Inc. (IAI), along with our academic and industry collaborators, proposes to develop a Scalable and Automated Tool to Analyze and Identify Dual Use Research of Concern from Scientific Publications: SAT-DURC. The proposed SAT-DURC integrates the following components: (1) data collection and integration module to prepare data for further analysis, (2) advanced analytics module to analyze and identify DURC potential, and (3) user interface to deliver results and annotations to users. IAIs big data technologies will be the foundation of all the proposed algorithms and architectures. The key innovation of the proposed approach is a novel combination of biomedical Natural Language Processing (NLP)-based pathogen and toxin identification incorporating customized ontologies and deep learning-based weakly-supervised experimental category classification, so that it can automatically detect DURC potential in scientific publications only requiring limited labeled data.

Intelligent Automation, Inc. (IAI), along with our collaborators, has successfully conducted a proof-of-concept study for a Scalable and Automated Tool to Analyze and Identify Dual Use Research of Concern from Scientific Publications: SAT-DURC. The key innovation of our proposed approach is a novel DURC identification system that combines the strength of natural language processing with machine learning, leverages the domain knowledge of subject matter experts and provides a user friendly interface. In Phase I, we have achieved satisfactory preliminary results for DURC identification of one bio-agent and one experimental category by applying advanced deep learning and Natural Language Processing based algorithms. The focus of the Phase II project will be to acquire a large dataset, enhance and extend the algorithms developed in Phase I to all 15 bio-agents and 7 experimental categories with the same level or better DURC identification accuracy, evaluate the algorithms with the large dataset, develop a user-friendly SAT-DURC software tool, and work with our commercial partners to transition this technology.