A Machine Learning Approach to Identify a Circulating MicroRNA Signature for Alzheimer Disease.

Background: Accurate diagnosis of Alzheimer disease (AD) involving less invasive molecular procedures and at reasonable cost is an unmet medical need. We identified a serum miRNA signature for AD that is less invasive than a measure in cerebrospinal fluid.

Methods: From the Oxford Project to Investigate Memory and Aging (OPTIMA) study, 96 serum samples were profiled by a multiplex (>500 analytes) microRNA (miRNA) reverse transcription quantitative PCR analysis, including 51 controls, 32 samples from patients with AD, and 13 samples from patients with mild cognitive impairment (MCI). Clinical diagnosis of a subset of AD and the controls was confirmed by postmortem (PM) histologic examination of brain tissue. In a machine learning approach, the AD and control samples were split 70:30 as the training and test cohorts. A multivariate random forest statistical analysis was applied to construct and test a miRNA signature for AD identification. In addition, the MCI participants were included in the test cohort to assess whether the signature can identify early AD patients.

Results: A 12-miRNA signature for AD identification was constructed in the training cohort, demonstrating 76.0% accuracy in the independent test cohort with 90.0% sensitivity and 66.7% specificity. The signature, however, was not able to identify MCI participants. With a subset of AD and control participants with PM-confirmed diagnosis status, a separate 12-miRNA signature was constructed. Although sample size was limited, the PM-confirmed signature demonstrated improved accuracy of 85.7%, largely owing to improved specificity of 80.0% with comparable sensitivity of 88.9%.

Conclusion: Although additional and more diverse cohorts are needed for further clinical validation of the robustness, the miRNA signature appears to be a promising blood test to diagnose AD.