Generating knowledge networks from phenotypic descriptions

摘要

Several computing systems rely on information about living beings, such as Identification Keys — artifacts created by biologists to identify specimens following a flow of questions about their observable characters (phenotype). These questions are described in a free-text format, e.g., “big and black eye”. Free-texts hamper the automatic information interpretation by machines, limiting their ability to perform search and comparison of terms, as well as integration tasks. This paper proposes a method to extract phenotypic information from natural language texts from biology legacy information systems, transforming them in an Entity-Quality formalism — a format to represent each phenotype character (Entity) and its state (Quality). Our approach aligns automatically recognized. Entities and Qualities with domain concepts described in ontologies. It adopts existing Natural Language Processing techniques, adding an extra original step, which exploits intrinsic characteristics of phenotypic descriptions and of the organizational structure of Identification Keys. The approach was validated over the FishBase data. We conducted extensive experiments based on a manually annotated Gold Standard set to assess the precision and applicability of the proposed extraction method. The obtained results reveal the feasibility of our technique, its benefits and possibilities of scientific studies using the extracted knowledge network.