Text mining applied to PubMed searches on Hemolytic Uremic Syndrome

CRISITINA IBARRA; RICARDO DORR; ROXANA TORIANO; CLAUDIA SILBERSTEIN

Rosario

Congreso; Sociedad Argentina de Fisiología. Reunión anual 2019; 2019

SAFIS

Human Hemolytic Uremic Syndrome (HUS) is characterized by the simultaneous development ofnonimmune hemolytic anemia, thrombocytopenia, and acute renal failure. Different causes lead to the syndrome, but the more frequent is the infection caused by Shiga toxin-producing Escherichia coli (STEC), present in food and water supplies. STEC causes human gastrointestinal infections and the developing of HUS in 15% of the cases. According to the World Health Organization, Argentina has the highest global incidence rate of HUS in children under five. HUS can cause death and is the leading cause of acute renal failure in pediatric patients. From the first publication in 1955 a great number of papers have contributed to the understanding of HUS. At the same time, with the exponential increase in the number of articles published each year on biomedical topics, it is raised as a necessity in science to build automated systems to extract information from them. Our hypothesis holds that text-mining on scientific databases offers a powerful tool to analyze behaviors, track tendencies and make predictions. To test, we have carried out an in-depth data-mining analysis on the results of a search on HUS of all publications indexed in MEDLINE up to 2018. Our main goal was to analyze the underlying text at the level of the descriptors used in searches and to discover information structures and nonexplicit (often hidden) patterns. Different informatic tools were applied: Knime Analytics Platform; Voyant tools and AntConc. We show the text mining results on a set from 7989 original articles with more than 3.2 x 106 words. We analyzed 5949 abstracts, 40851 authors and 6191 affiliations to obtaining new valuable information in the study of HUS. Also, we worked on bag of words to analyze temporal frequencies and to do forecasting, and we applied unsupervised computing techniques in topic extraction. As a conclusion, we believe that text mining is an important tool to enriching understanding and promoting disease prevention.Suporting by PUE 2017 #22920170100041CO - CONICET; UBACyT 2017-2020 #20020170100733BA UBA; NVIDIA Corporation for donation of the Titan Xp GPU used for our researc