Classifying Usual Interstitial Pneumonia in Patients with Interstitial Lung Disease Using Machine Learning on High-Dimensional Transcriptional Data

• Activity Number: 693
• Type: Contributed
• Date/Time: Thursday, August 13, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Medical Devices and Diagnostics
• Author(s): Su Yeon Kim* and James Diggans and Dan Pankratz and Jing Huang and Moraima Pagan and Nicole Sindy and Yoonha Choi and Giulia C. Kennedy
• Companies: and Veracyte, Inc. and Veracyte, Inc. and Veracyte, Inc. and Veracyte, Inc. and Veracyte, Inc. and Veracyte, Inc. and Veracyte, Inc.
• Keywords: Gene expression ; microarrays ; RNA transcription ; classification ; algorithm ; surgical lung biopsies

Abstract

Idiopathic pulmonary fibrosis (IPF) is a deadly disease that involves progressive lung scarring and results in respiratory failure. Diagnosis is challenging as other interstitial lung diseases (ILDs) share similar histopathological characteristics. Usual interstitial pneumonia (UIP) pattern, a hallmark of IPF, is essential for diagnosing it. Our goal is to develop a genomic classifier that distinguishes UIP from other ILDs on surgical lung biopsies (SLB). 125 SLBs were collected from 86 ILD patients; SLB pathology diagnoses were reached by an expert panel: 58 are UIPs, 67 are non-UIPs. RNA expression for 33,297 transcripts was measured on microarrays. A classifier was trained on 77 samples and tested on 48. A subset of 36 samples was subjected to RNA sequencing (RNASeq), and a classifier trained on RNASeq was evaluated by cross-validation (CV). The microarray classifier achieves high test performance: AUC=0.94, specificity=0.92 and sensitivity=0.82. The RNASeq classifier achieves high CV performance with AUC=0.9. Our results demonstrate the feasibility of developing a genomic signature that predicts UIP pathology, an important step towards accurate diagnosis of IPF.