Go to The Journal of Clinical Investigation
  • About
  • Editors
  • Consulting Editors
  • For authors
  • Publication ethics
  • Transfers
  • Advertising
  • Job board
  • Contact
  • Current issue
  • Past issues
  • By specialty
    • COVID-19
    • Cardiology
    • Immunology
    • Metabolism
    • Nephrology
    • Oncology
    • Pulmonology
    • All ...
  • Videos
  • Collections
    • Resource and Technical Advances
    • Clinical Medicine
    • Reviews
    • Editorials
    • Perspectives
    • Top read articles
  • JCI This Month
    • Current issue
    • Past issues

  • Current issue
  • Past issues
  • Specialties
  • In-Press Preview
  • Editorials
  • Viewpoint
  • Top read articles
  • About
  • Editors
  • Consulting Editors
  • For authors
  • Publication ethics
  • Transfers
  • Advertising
  • Job board
  • Contact
Machine learning implicates the IL-18 signaling axis in severe asthma
Matthew J. Camiolo, … , Anuradha Ray, Sally E. Wenzel
Matthew J. Camiolo, … , Anuradha Ray, Sally E. Wenzel
Published September 30, 2021
Citation Information: JCI Insight. 2021;6(21):e149945. https://doi.org/10.1172/jci.insight.149945.
View: Text | PDF
Research Article Immunology Pulmonology

Machine learning implicates the IL-18 signaling axis in severe asthma

  • Text
  • PDF
Abstract

Asthma is a common disease with profoundly variable natural history and patient morbidity. Heterogeneity has long been appreciated, and much work has focused on identifying subgroups of patients with similar pathobiological underpinnings. Previous studies of the Severe Asthma Research Program (SARP) cohort linked gene expression changes to specific clinical and physiologic characteristics. While invaluable for hypothesis generation, these data include extensive candidate gene lists that complicate target identification and validation. In this analysis, we performed unsupervised clustering of the SARP cohort using bronchial epithelial cell gene expression data, identifying a transcriptional signature for participants suffering exacerbation-prone asthma with impaired lung function. Clinically, participants in this asthma cluster exhibited a mixed inflammatory process and bore transcriptional hallmarks of NF-κB and activator protein 1 (AP-1) activation, despite high corticosteroid exposure. Using supervised machine learning, we found a set of 31 genes that classified patients with high accuracy and could reconstitute clinical and transcriptional hallmarks of our patient clustering in an external cohort. Of these genes, IL18R1 (IL-18 Receptor 1) negatively associated with lung function and was highly expressed in the most severe patient cluster. We validated IL18R1 protein expression in lung tissue and identified downstream NF-κB and AP-1 activity, supporting IL-18 signaling in severe asthma pathogenesis and highlighting this approach for gene and pathway discovery.

Authors

Matthew J. Camiolo, Xiuxia Zhou, Qi Wei, Humberto E. Trejo Bittar, Naftali Kaminski, Anuradha Ray, Sally E. Wenzel

×

Figure 3

Machine learning validation of a 31-gene signature for patient clustering.

Options: View larger image (or click on image) Download as PowerPoint
Machine learning validation of a 31-gene signature for patient clusterin...
(A) Receiver operating characteristics (ROC) curve of a sparse-partial least squares discriminant analysis (sPLS-DA) model training for k-means cluster prediction on the SARP cohort. ROC curves were calculated as one class versus the others using 5 fold-validation on the original training set. Reported AUC are based on comparison of predicted scores of one class versus the others. (B) Circle plot demonstrating concordance of assignment between original SARP clustering and 31-gene solution. (C) Heatmap of expression of the genes included in clustering of the IMSA validation cohort with patient cluster or clinical disease severity as indicated in row sidebar. (D) GSEA results for indicated transcription factor target datasets in IVC4 vs remaining IMSA participants (REST). (E) Box plot of FEV1 across IMSA clusters with P value from Kruskal-Wallis testing. Error bars represent medians, with bounds of boxes representing IQR and whiskers representing 1.5× the upper or lower IQR. (F) Elastic net–predicted (EN-predicted) FEV1 based on gene expression versus measured FEV1% predicted in SARP or IMSA. Grayed area indicates the 95% confidence bounds around a linear regression model comparing the 2. Spearman’s ρ and P value are shown. (G) Graphical representation of EN modeling determinants of lung function (FEV1). Coefficients from SARP or IMSA are plotted in order of ascending value from left to right, with distance from the hashed line indicating magnitude of contribution to the model. Blue coloration of transcript ID denotes a negative coefficient, and red indicates positive. Asterisk in plot areas denote IL-18R1 in either data set.

Copyright © 2023 American Society for Clinical Investigation
ISSN 2379-3708

Sign up for email alerts