Abstract 19: Machine Learning Approach Identifies a Pattern of Gene Expression in Peripheral Blood Which Can Accurately Detect Ischemic Stroke

Abstract

Objective: The identification of stroke-associated biomarkers represents a means by which prehospital triage could be expedited to increase the probability of successful intervention. Thus, the objective of this work was to use high-throughput transcriptomics in combination with basic machine learning techniques to identify a pattern of gene expression in peripheral whole blood which could be used to identify acute ischemic stroke (AIS) in the acute care setting. Methods: A two-stage study design was used which included a discovery cohort and an independent validation cohort. In the discovery cohort, peripheral whole blood samples were obtained from 39 AIS patients upon emergency department admission, and from 24 neurologically asymptomatic controls. Microarray was used to measure the expression of over 22,000 genes and a pattern recognition technique known as genetic algorithm k-nearest neighbors (GA/kNN) identified a pattern of gene expression that optimally discriminated between AIS and controls. In an independent validation cohort, the gene expression pattern was tested for its ability to discriminate between 39 AIS patients and each of two different control groups, one consisting of 30 neurologically asymptomatic controls, and the other consisting of 15 stroke mimics, with gene expression levels being assessed by qRT-PCR. Results: In the discovery cohort, GA/kNN identified ten transcripts (ANTXR2, STK3, PDK4, CD163, MAL, GRAP, ID3, CTSZ, KIF1B, and PLXDC2) whose coordinate pattern of expression correctly identified 98.4% of subjects (97.4% sensitive, 100% specific). In the validation cohort, the same 10 transcripts correctly identified 95.6% of subjects when comparing AIS patients to asymptomatic controls (92.3% sensitive, 100% specific), and 96.3% of subjects when comparing AIS patients to stroke mimics (97.4% specific, 93.3% sensitive). Conclusion: These results demonstrate that a highly accurate RNA-based companion diagnostic for AIS is plausible using a relatively small number of markers. The pattern of gene expression identified in this study shows strong diagnostic potential, and warrants further evaluation to determine true clinical efficacy.