1681. Female Urinary Metagenomic Analysis and Natural Language Processing Enhances the Infectious Diagnostic Yield in Precision Medicine. (31st December 2020)

• Record Type: Journal Article
• Authors: Valencia, C Alexander; Baugher, David; Larsen, Alexander; Chen, Alvin; Icenhour, Crystal
• Is Part Of: Open forum infectious diseases. Volume 7:Number 1(2020) Supplement
• Page Start: S825
• Page End: S825
• Publication Date: 2020-12-31
• Subjects: Communicable diseases -- Periodicals; Medical microbiology -- Periodicals; Infection -- Periodicals; 616.9
• DOI: 10.1093/ofid/ofaa439.1859
• Languages: English
• ISSNs: 2328-8957

Abstract

Abstract: Background: In this study, we assessed the diagnostic yield of metagenomics urine sample testing in patients with urological symptoms. Methods: We conducted metagenomic analysis on 69 consecutive unbiased female patients, by sequencing their DNA using the KAPA HyperPlus library construction with next-generation sequencing (Nextseq500, Illumina) and reads were analyzed using Xplore-Patho®, an analytical system that permits the detection of 37, 000+ microorganisms, including over 12, 000 known pathogens, and examined report summaries written by infectious disease experts to obtain a diagnostic yield. In addition, infectious disease expert analysis was contrasted with a natural language (NLP) pathogen detection system to investigate its accuracy. Results: In the expert data summaries, a total of 95% of the patients tested had at least one pathogen identified by metagenomics as a potential explanation of their urological symptoms and these results were binned into four categories: 1) 51% of infection likely, 2) 4% of infection possible, 3) 26% of low-grade infection likely and 4) 14% of low-grade infection possible. Data from healthy controls was used in conjunction with an NLP pathogen detection pipeline and compared to infectious disease expert summaries. The NLP pathogen algorithm detected that at least 97% of samples had one pathogen which was more than 5 standard deviations from the abundance of that pathogen in healthy controls, and least 84% had 2 or more pathogens. These diagnostic percentages were consistent with the infectious disease expert summaries. The NLP algorithm had access to a large database derived from PubMed articles and it was found that several relevant uropathogens were not mentioned in report summaries. For example, one well-documented uropathogen was present in 13 samples, but was not mentioned in any report summaries. Conclusion: In conclusion, this study demonstrated the high diagnostic yield in females with urological symptoms following metagenomic analysis and the ability of NLP to enhance the sensitivity of reportable pathogens. Disclosures: All Authors : No reported disclosures