Machine Learning Optimization: Defining Exposome-Metabolome Associated Aerodigestive Disease

Details

• Personal Author:
  Crowley G ; Grunig G ; Kwon S ; Liu M ; McRitchie S ; Nolan A ; Podury S ; Prezant DJ ; Rushing B ; Sumner S
• Description:
  RATIONALE Particulate matter(PM) exposure contributes to 7-million deaths annually. Aerodigestive complications include airway hyperreactivity(AHR), gastroesophageal reflux disease(GERD) and premalignant Barrett's Esophagus(BE). While studies are limited in clarifying the pathophysiological link between AHR, GERD and BE, discovery of biomarkers may yield biologically plausible pathways and identify populations needing targeted intervention. Detecting complex associations in high-dimensional datasets is a unique challenge. Machine learning(ML) is important in the analysis of untargeted metabolomics. Our work focused on improving the speed, precision, and accuracy of data-mining through use of random forests(RF) with gradient-boosted tree modeling to identify metabolite classifiers of AHR, GERD and BE. METHODS World Trade Center(WTC)-PM exposed firefighters with normal pre-exposure lung function were defined as cases of AHR by electronic medical record(EMR), PC20<16mg/ml, and/or positive bronchodilator-response; GERD and BE status determined by EMR or endoscopy biopsy diagnoses. Controls were defined as not meeting any case criteria and were members of the 10%-randomly selected cohort controls. Metabolite serum profiles were analyzed by liquid chromatography-mass spectrometry and peaks were matched using in-house and public libraries. Machine Learning. RF of the untargeted, filtered, normalized metabolome(n=14,969 peaks) yielded a refined metabolite profile of the top 5% by mean decreased accuracy(MDA), and were included in a gradient-boosted tree model(xgboost,R-Project) to build a case classifier. The final model was determined by random hyperparameter space search, maximizing cross-validated AUCROC. RESULTS The refined profiles for all cases of AHR, GERD, and BE are shown with values of log2 (median-fold-change). AHR cases had decreased mesobilirubinogen compared to controls(5-fold cross-validated AUCROC=0.95;15% classification error(CE)),Fig-1A. GERD identified metabolites related to estrogen/testosterone metabolism: 11-oxo-androsterone glucuronide is a metabolite of testosterone; 2,3-bis(4-hydroxyphenyl) propionitrile is an agonist for estrogen receptor beta, 4-hydroxyandrostenedione glucuronide is a metabolite of the aromatase inhibitor 4- hydroxyandrostenedione. No sex hormone metabolites were identified as key metabolites of BE. Acetaminophen and moxifloxacin were key exposures in GERD and BE, respectively. No overlapping metabolites identified cases of GERD and BE. Final models AUCROC=0.88;19.5% CE for GERD and an AUCROC=0.93; CE=19.3%; for BE. CONCLUSIONS Our findings included differential expression of mesobilirubinogen, a metabolite of biliverdin. Heme oxygenase (HO)-1 metabolizes heme into biliverdin. Sex hormones and exogenous therapeutics were linked to GERD and BE. Transient receptor potential(TRP)V1 gene expression is altered in esophageal mucosa of reflux patients and TRPV1 is an acetaminophen target. Metabolic profiling provides a contemporaneous snapshot of an organism's physiology and may identify plausible pathways of disease that could be attenuated pharmacologically. [Description provided by NIOSH]
• Subjects:
  Biomarkers Occupational Health Particulate Matter Safety September 11 Terrorist Attacks
• Keywords:
  Gastroesophageal Reflux Disease GERD Machine Learning Particulates World Trade Center WTC
• ISSN:
  1073-449X
• Document Type:
  Text
• Funding:
  Cooperative Agreement
• Genre:
  Abstract or Summary
• Place as Subject:
  New York ; North Carolina ; OSHA Region 2 ; OSHA Region 4
• CIO:
  National Institute for Occupational Safety and Health (NIOSH)
• Topic:
  Workplace Safety & Health
• Location:
  North America
• Volume:
  209
• NIOSHTIC Number:
  nn:20070822
• Citation:
  Am J Respir Crit Care Med 2024 May; 209(Abstract Issue):A3894
• Federal Fiscal Year:
  2024
• Performing Organization:
  New York University School of Medicine
• Peer Reviewed:
  False
• Start Date:
  20210701
• Source Full Name:
  American Journal of Respiratory and Critical Care Medicine
• Supplement:
  Abstract Issue
• End Date:
  20260630
• Collection(s):
  National Institute for Occupational Safety and Health
• Main Document Checksum:
  urn:sha-512:657c16cafdbc01a6467e2ca4a6c1ffc91b0ca37908974dfd8873b9230f9cf72e4d3b8631f5954228d57e5cceefd824253910c2bc29abf20a19b56b86167c8865
• Download URL:
  https://stacks.cdc.gov/view/cdc/208971/cdc_208971_DS1.pdf
• File Type:
  [PDF - 267.35 KB ]