Abstract

Literature-based Discovery (LBD) uses information from explicit statements in literature to generate new knowledge and can thus facilitate hypothesis testing and generation from publications to accelerate scientific research. Existing methods, however, use methodologies which are inadequate for capturing the complex information available in scientific literature and are prone to proposing spurious or low-quality discoveries. Recent advances in NLP allow for deep textual analysis to obtain a wide coverage of information in text and adapt to recognising new entities. Similarly, recent advances in graph processing have made it possible to do in-depth analysis on information represented as graphs to facilitate knowledge discovery. Both advances utilise neural networks extensively. This work used neural networks to advance LBD by: improving biomedical NER using multi-task learning; improving knowledge discovery from biomedical graphs using link prediction; and improving the ranking of published discoveries by scoring the strength of connection paths. Excitingly, the latter approaches outperformed those used by the state-of-the-art LION LBD tool. These results show that it is feasible to use neural networks to improve this increasingly necessary task and that neural biomedical knowledge discovery is potent, operational and a potentially rich field for further study.