SBIR Phase II: Serendipitous Search System Using Lateral Analogy to Match Potential Solutions to Unmet Needs:Feasibility Study Based on Screening Approved Drugs for Repurposing

The broader impact/commercial potential of this project is to accelerate the pace of research and development to enable more rapid deployment of technologies into commercial / industrial contexts. In many fields, information is expanding at such an exponential rate that finding relevant results to technical knowledge searches is increasingly difficult. Further, content is expanding so fast that most fields are rapidly forming sub-disciplines, leading to the ?silo-ing? of different knowledge sub-domains, a clear challenge to both academia and industry. We need ever better ways to organize and present information to users. There are disadvantages of the current search engines, mostly relating to excessive similarity in search results. Further, while these engines present information relating to a known search target, they are less effective at presenting unexpected results for information that a user has never heard of but that would be useful. What is therefore needed is an exploration system giving searchers a strong serendipitous element with a maximum likelihood of results from diverse, unexpected, and potentially provocative sources. This will break down silos by providing a rapid, relevant means for knowledge-transfer between different disciplines, fostering interdisciplinary innovation. This system has been designed to provide a means for systematic, automated discovery. This Small Business Innovation Research (SBIR) Phase 2 project is focused on optimizing and scaling a serendipitous document search system for repurposing technologies by analogy into lateral fields. Both by sub-parsing discrete content into ontologically separable entities, such as capability, characteristic, and composition, and by comparatively assessing certain of these attributes between such entities, the attribute relatedness of these entities can be used to drive their self-assembly into related attribute networks. This approach provides a significant value proposition for drug repurposing, which is the current focus of this project. To scale the pair-wise comparison and network assembly of millions of documents, a map-reduce based text-processing framework will be developed so that massively parallel computations can be carried out in a time- and costefficient manner. A distributed search engine technology will be deployed to enable rapid querying of the emerging document relationship network. A series of machine learning algorithms will then be used to determine potentially hidden structural architectural features within the document relationship network. Machine learning will elucidate the nature of the relationships in drug networks through analyses of inter-node relationships and sub-graph motifs (termed ?innovation motifs?). Documents including U.S. patents and scientific papers will be processed in the system.