Sensorineural hearing loss (SNHL) is the most common neurodegenerative disease, estimated to affect over 432 million adults and 34 million children worldwide.1 Despite this prevalence, clinical testing does not yield a cell- or molecular-based identification of the underlying etiology of hearing loss, making it difficult to develop pharmacological or molecular treatments. A key to improving diagnosis and treatment is the development of reliable biomarkers for the myriad of active inner ear disorders. Micro-RNAs (miRNAs) are short ribonucleic acids that regulate a variety of cellular processes, including post-transcriptional gene regulation. They have been identified in all body fluids, and gained significant momentum within the literature over the recent decades because of their prominent role in various disease states, from cancer to Alzheimer's.2 Interestingly, miRNAs have been recognized as a promising diagnostic and prognostic marker for otherwise difficult-to-diagnose neurodegenerative diseases such as Alzheimer's and Parkinson's disease.3
Our group recently demonstrated that miRNA profiling within the inner ear using human perilymph is a feasible and safe method that can offer significant insights into inner ear disorders on cellular and molecular levels.4 In our search for hearing-specific biomarkers, we discovered that various inner ear disorders exhibited distinct disease-specific miRNA expression profiles. Along similar lines, multiple studies have shown that miRNAs play a significant role in sudden sensorineural hearing loss and inner ear development.5 However, some challenges in analyzing miRNA data from the inner ear are the immense and variable expression patterns across different disease states.
Machine learning (ML) offers a solution to this challenge because of its ability to quickly analyze and learn from large amounts of information. ML is a sub-discipline of artificial intelligence that borrows from multiple disciplines, including mathematics, statistics, and computer science. ML is also able to recognize patterns and make predictions based on what it has learned. While statistics is focused on inferences related to causation and how a system of components relate to each other, ML is a novel way of analyzing data in that it makes predictions based on large sets of data.
In our work, we used ML to build disease-specific algorithms to predict various degrees of SNHL in different inner ear pathologies based on perilymph-derived miRNA expression profile alone.6 We collected 2 to 5 uL of perilymph from patients whose inner ears were opened as part of a procedure (cochlear implantation and stapedectomy). We analyzed the perilymph using the Affymetrix GeneChip miRNA 4.0 microarray (Fig. 1). All patients underwent pure tone audiometry (PTA) prior to their respective procedures. We then analyzed the miRNA dataset unique to inner ear pathologies using a supervised machine learning classification model and considered multiple decision models, including multiclass decision forest, decision jungle, logistic regression, and neural networks. We created the model using a 70/30 split, where 70 percent of the patients were used to construct the model and the other 30 percent were used to test the ML model. The permutation feature of importance in ML allows it to understand which component and at what weighted value that component was used to create the model. In this case, the permutation feature of importance listed the key miRNA that the algorithm used to make the diagnosis. After constructing the model, we introduced a blind miRNA sample to determine if ML could make an accurate diagnosis based on the perilymph miRNA expression profile alone.
CONDUCTIVE HEARING LOSS VS. SNHL
We first compared the miRNA expression profiles in patients who had stapedectomy and represented the pure conductive hearing loss (CHL) group with patients who had cochlear implantation and represented the SNHL group. Both the decision forest and logistic regression ML models were able to distinguish SNHL from CHL with 100 percent accuracy. Conversely, the decision jungle and neural network ML models were able to distinguish SNHL from CHL with 80 percent accuracy.
SNHL WITH VS. WITHOUT RESIDUAL HEARING
We then compared the miRNA expression profiles of patients who had cochlear implantation with various degrees of residual hearing. These patients were classified as either having residual hearing (cochlear implant patients with PTA <80 dB) or no residual hearing (cochlear implant patients with PTA > 80 dB). All four ML models were able to differentiate between cochlear implants with and without residual hearing with 100 percent accuracy.
The permutation feature of importance also offers a novel way to analyze miRNA that may be critical in various inner ear pathologies. It gives a weighted score to the miRNAs that are critical in developing the decision-making algorithm and influence the model. Using the Ingenuity Pathway Analysis software (Qiagen Bioinformatics), the key miRNAs were analyzed alongside a human cochlea cDNA library. Known and highly predicted miRNA cochlear mRNA interactions were identified and mapped out to understand the different regulatory pathways.6
In this study, we demonstrated that miRNAs not only show ear disease-specific profiles, but an miRNA expression profile can also be used to diagnose various degrees of SNHL. This methodology provides an understanding of inner ear pathology on a molecular level, and offers a novel method to diagnose and prognose patients with active inner ear disease in a way not previously possible. Similar to how patients undergo a lumbar puncture to diagnose meningitis, one could theoretically undergo a round window tap to diagnose, prognose, and monitor therapeutic interventions for various inner ear disorders. While this does offer an exciting prospect for stratification of patients with inner ear disease, multiple safety and validation studies will need to be performed.
2. Vidigal JA, Ventura A The biological functions of miRNAs: lessons from in vivo studies Trends in cell biology 2015 25 3 137–147
3. Burgos K, Malenica I, Metpally R, et al Profiles of extracellular miRNA in cerebrospinal fluid and serum from patients with Alzheimer's and Parkinson's diseases correlate with disease status and features of pathology PLoS One 2014 9 5 e94839
4. Shew M, Warnecke A, Lenarz T, Schmitt H, Gunewardena S, Staecker H Feasibility of microRNA profiling in human inner ear perilymph Neuroreport 2018 29 11 894–901
5. Li Q, Peng X, Huang H, Li J, Wang F, Wang J RNA sequencing uncovers the key microRNAs potentially contributing to sudden sensorineural hearing loss Medicine 2017 96 47 e8837
Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.
6. Shew M, New J, Wichova H, Koestler DC, Staecker H Using Machine Learning to Predict Sensorineural Hearing Loss Based on Perilymph Micro RNA Expression Profile Scientific reports 2019 9 1 3393