Ranking Literature from the Network of Drug-Disease Association through Multi-Layered Semantic Model

This thesis presents the design and impementation of a system to find the top ranked relevant Medline citations with hyperlinks. The iput to the systm is a network of semantically related drug-diseases associations obtained by the DDNet. A fully functional system can be used for generation/validation of hypothesis or to find most relevant literature based on user input in birdging gap between the generation and utilzation of biomedical literature. Such a system may also have an impact in identification of drug targets and may open new avenues for drug repositioning in clinical and translational research. Multi-Layered LSA model is implemented to filter the voluminous publications and find top ranked literature. In particular, a Local Latent Semantic Analysis (LLSA) with query-based sampling of abstracts was used to build model that is relatively free from garbage-in garbage-out syndrome . The current implementation ensures that the system is efficient, scalable and relatively free from systemic bias. Also, the concept of mapping ontologies was adopted to develop best suited dictionary to find the domain specific crisp associaions and relevant results. A reverse ontology mapping was used to create a network from semantically relevant associations. In addition, a web service application was developed to query the system and visualize the computed network of associations along with their most relevant publications\u27 links in a way that is easy to interact. A pilot study was conducted to evaluate the performance of system using both subjective and objective measures