Development of methods for automatic extraction of knowledge from texts of scientific publications for the creation of a knowledge base Solanum tuberosum

Currently there are hundreds of scientific journals that publish research results in various fields of plant biology and agrobiology. Hundreds of thousands of international patents contain a variety of information on agricultural biotechnology. The number of articles and patents is increasing over time in an exponential progression. For example, there are more than 1.5 million publications devoted to the study of Solanum tuberosum that is one of the most important crops in the world. Analysis of such huge number of experimental facts presented in text sources (scientific publications and patents), requires the use of automated methods for knowledge extraction (text-mining). Intelligent automatic text analysis techniques are already widely used in biology and medicine to extract information about the properties and functions of molecular genetic objects. Unlike search engines such as Google, Yandex and others, that search documents by keywords, such text-mining methods are aimed at the automatic extraction of knowledge presented in the documents, knowledge integration and formalization according to the defined ontology. Among the known systems for intelligent knowledge extraction from scientific publications STRING, LMMA, ConReg, GeneMania and others can be listed. For the first time in Russia, we have previously developed a system, named ANDSystem, for automatic intelligent knowledge extraction in biomedicine. ANDSystem contains more than 10 million facts about molecular-genetic interactions extracted from more than 25 million scientific publications. For knowledge extraction in ANDSystem, specially developed semantic and linguistic rules are used for recognition of interactions between biological objects such as, proteins, genes, metabolites, drugs, miсroRNA, biological processes, diseases and others in natural language texts. However, the problem of development of methods for automatic knowledge extraction from the texts in plant biology, agrobiology and agrobiotechnology remains still unsolved and has a high relevance. The aim of this work was to adapt the methods of automatic knowledge extraction, presented in ANDSystem, to the field of crop production and to create on this basis a SOLANUM TUBEROSUM knowledge base, containing information on genetics, markers, breeding and selection of potatoes, its pathogens and pests, storage and processing technologies and others. The knowledge base ontology contains dictionaries, corresponding to more than 20 types of objects, including molecular genetic objects (proteins, genes, metabolites, microRNA, biological processes, biomarkers, etc.), potato varieties and their phenotypic traits, diseases and pests of potato, biotic and abiotic environmental factors, biotechnologies of cultivation, processing and storage of potato, and others. Also, the ontology contains more than 25 types of interactions that describe various relationships between the above listed objects, including molecular interactions, regulatory events and associative links. More than 5 thousand semantic templates were created to extract information about the interactions. The accuracy and recall of knowledge extraction by the developed method were assessed with the expert manual analysis of the text corpus and reached more than 65 % and 70 %, respectively. The full-scale version of the knowledge base SOLANUM TUBEROSUM will be created on the basis of the developed approaches.