A platform for rapid knowledge base construction and hypotheses exploration using extractive literature search

As researchers struggle to keep track of the published scientific literature`s growing extent, we aim to provide an alternative way to represent and review the current state of knowledge. We developed a platform for knowledge base construction, assisted by Natural Language Processing (NLP) for robust, hypothesis-driven curation of defined relation instances between entities of interest.

Syntactic queries were applied to produce lists of biomedical entities from mined text according to relevant classes, and explicit relation instances between them were automatically captured. A dedicated user interface facilitates the evidence annotation and generation of a comprehensive overview, provided with an interactive network tree to explore complex hypotheses in a combinatorial manner.

Using this method, we constructed a comprehensive ‘Cell-Specific Targeted Drug Delivery’ knowledge base, containing thousands of annotated relation instances between surface targets, ligands, cancers, cell types, biomaterials and drugs; These are now openly available to facilitate NLP’s implementation in biomedicine and could provide an alternative approach tasking conventional literature exploration, curating references for systematic literature reviews and performing meta-analyses.