Machine Learning Model Identifies Genomic Variation in Noise-Induced Hearing Loss via Whole-exome Sequencing

doi:10.21203/rs.3.rs-1829497/v2

Download PDF

Research Article

Machine Learning Model Identifies Genomic Variation in Noise-Induced Hearing Loss via Whole-exome Sequencing

https://doi.org/10.21203/rs.3.rs-1829497/v2

This work is licensed under a CC BY 4.0 License

Version 2

posted

You are reading this latest preprint version

Noise-induced hearing loss (NIHL) is related to noise exposure and whether humans suffered from NIHL is also related to different sensitivity to noise. Whole exome sequencing (WES) can detect the genetic mutations of NIHL population so as to find the susceptibility of different people from the genetic level. WES can identify contributing genes, not only find NIHL susceptibility genes, but also divide patient cohort into different subtypes. We conducted a case-control (n = 78, n = 35, respectively) study to find the susceptibility genes of NIHL through WES and stratified the cases into subtypes, so as to achieve the purpose of precision medicine. 73 gene variants were used to train the unsupervised machine learning model which separated the case from the control group and divide the population into two subtypes. Supervised machine learning methods compared the above constructed subgroup classification models. This is the first study to link NIHL with gene variants and has produced a new potential classification method that can be applied to precision medicine in people’s different sensitivity to noise. These observations deserve further study on independent and larger clinical samples in order to provide new information for future research on the susceptibility of NIHL.

noise-induced hearing loss

whole exome sequencing

machine learning

prediction model

Hearing is one of our most developed senses. It depends on the appropriate function of the auditory system and plays an important role in people's social communication and learning ability. However, many factors such as environmental stimulation, aging and chemicals can lead to severe inner ear hearing loss. Noise-induced hearing loss is a kind of disease in which people's hearing has gradually decreased after long-term exposure to noise. It is easy to get hearing loss in the environment of long-term exposure to noise, such as factories and the army. Noise is one of the health hazards in factories, armies or communities and the hearing loss caused by noise accounts for a large proportion. With the gradual refinement of modern occupational classification, some special occupations are exposed to noise for a long time, with dense frequency, and the problem of noise induced hearing loss is becoming more and more prominent. Noise-induced hearing loss is the second most common sensorineural hearing loss after age-related hearing loss[1]. Globally, it is estimated that 16% − 24% of hearing loss is related to noise[2].

Genetic variations can lead to hearing loss to a large extent and can also lead to people’s different sensitivity to noise[3]. In all hearing loss patients, about half of severe hearing loss cases have genetic changes[4, 5]. Noise-induced hearing loss is not only caused by genetic factors, but also by environmental factors, and is related to the intensity and duration of noise[6]. However, the susceptibility of different people to noise induced hearing loss varies greatly, which may be related to their different genetic background[7, 8]. The genetics of noise induced hearing loss has been studied in animals and several knockout mice SOD1-/-, GPX1-/-, PMCA2-/- and CDH23+/- have been shown to be more sensitive to noise than their wild-type[8–12]. Hundreds of genes involving complex traits have also been found in mouse genome-wide association study (GWAS)[13]. Lavinsky j et al. Confirmed that NOX3 is a key protective gene for noise induced hearing loss through GWAS in mice[14]. In the population exposed to noise, there are many studies on single nucleotide polymorphisms (SNPs) which have used case-control studies to select susceptible individuals with 3, 4 or 6 kHz hearing loss for SNP research. These studies have found many NIHL susceptibility genes, such as SOD2, KCNQ1, PCDH15, HSP70, which are related to the mechanisms of oxidative stress and potassium circulation[8]. However, the past low-throughput analysis methods still have limitations in analyzing NIHL susceptibility genes.

Whole exome sequencing (WES) is a genomic analysis method that uses sequence capture technology to capture and enrich the exon region of the whole genome for high-throughput sequencing. The comprehensive analysis of the whole exon is helpful to detect the coding region mutation related to the disease, so as to study the impact of DNA variation on the normal biological mechanism and disease pathogenesis. WES is widely used in identifying deafness genes in families[15, 16]. In other diseases, it is also widely used in identifying susceptibility genes in sporadic cases. However, few people use WES in noise induced hearing loss. WES is helpful to find the susceptibility genes of noise induced loss by observing the hearing loss in people exposed to noise for a long time. Genome wide search can increase the ability to find susceptible genetic variation, so as to break through the limitations of SNP low throughput research.

In the big data of WES sequencing results, the analysis of susceptibility genes by machine learning has been widely used[17, 18]. Machine learning algorithms can be divided into supervised learning and unsupervised learning[19]. The application of machine learning in medical research to solve patients' problems in medicine has been a recent trend. And the method of machine learning is also very suitable for dealing with large high-dimensional data sets, such as genome data. Establishing a reasonable machine learning gene analysis model of noise induced hearing loss is not only the basis for exploring the etiology and treatment of noise induced hearing loss on the basis of genetics, but also the basis for noise protection and screening in special posts or specific populations.

This study will use unsupervised machine learning algorithm and supervised machine learning algorithm to analyze the WES sequencing results of NIHL. To our knowledge, this is the first study on WES machine learning analysis of NIHL susceptible population.

2.1. Participants and Inclusion Criteria

This study was conducted in accordance with the ethical standards of the Helsinki Declaration and the national and international norms and has been approved by the author's institutional ethics review board. In an experimental field exposed to impulse noise, we chose 78 experimental personnel for more than 1 year with pure tone audiometry showing hearing loss as NIHL group and 35 experimental personnel who have been exposed to noise for more than 3 years but have no hearing loss were selected as the control group. Before exposure to noise, all experimental personnel had normal hearing.

The inclusion criterion of the NIHL group is that in the frequency of 250 Hz to 6 kHz, as long as the hearing threshold of a frequency is greater than 25 dB. The inclusion criterion of the control group is that no frequency of 250 Hz to 6 kHz is greater than 25 dB. The study was approved by the institutional review committee of the PLA strategic support force characteristic medical center.

2.2. Epidemiological survey and hearing test

We investigated their age, gender, years of exposure to noise, and asked about the history of other diseases that may damage their hearing before the hearing test. Pure tone audiometry (Astera, Denmark) was used to test the subjects' auditory function in November 2019. Then compare the hearing experimenters between two groups at 0.25 kHz, 0.5 kHz, 1 kHz, 2 kHz, 3 kHz, 4 kHz, 6 kHz, high frequency threshold, and speech recognition threshold in pure tone audiometry under the condition of long-term exposure to impulse noise.

2.3 Whole Exome Sequencing

All participants collected peripheral blood and isolated peripheral blood DNA. Using the whole exome liquid phase capture technology, human whole exon region DNA was highly and specifically enriched. After enrichment, high-throughput and high-depth sequencing was carried out on HiSeq X10 or NovaSeq platform. PE150 sequencing was performed using Illumina X10 or NovaSeq 6000 platform (iGenetech, Beijing, China). The captured exons were double-terminal sequenced, the average read length was 2 × 150 bp and a sequencing depth was 100 ×. High quality cleaning reads were obtained from raw data through strict filtering. After the data is obtained by sequencing and filtering, the information analysis process is carried out based on the reference genome of UCSC (hg19, 2009). The analysis type is single sample whole exon analysis. BWA software compares the sequenced data with the reference genome, and performs sequencing and repeat sequence removal at the same time to obtain BAM result file. Based on the BAM results compared with the genome reference sequence, we used SAMtools, GATK and other software to find SNP and indel, and then annotated the SNP and indel sites with ANOVAR software to determine the gene information, function information and harmfulness corresponding to the mutation sites[20–22]. SNP genomic variation was annotated and analyzed using external databases to determine genomic location, mutation frequency, genotype heterozygosity, and the functional pathway of variation. The screening criteria of SNP include: 1) exon region; 2) nonsynonymous mutation; 3) the allele frequency was less than 0.01 in the East Asian group of ExAC, genomAD and 1000-genome database; 4) In SIFT, GERP, Polyphen2, Mutation Taster and CADD software, the prediction of mutation information is harmful.

2.4 Patient stratification and prediction of noise-induced hearing loss subtypes

In order to identify the specific mutations that occurred in the NIHL group compared with the control group, the variations detected in at least 5% of the samples in NIHL group but not in any samples in control group were also selected as the NIHL group specific mutation. The variable processing and prioritization, as well as the above statistical analysis, were done in the Python3.

Patients in the NIHL group were stratified based on their genotype. In order to find different NIHL subtypes in the study population, unsupervised hierarchical clustering-similarity and genome distance values calculated using the Jaccard coefficient[23] - was used to group patients according to the genome variation spectrum. The subtypes of hearing loss were generated by cluster analysis and a random forest model was generated to distinguish the subtypes of NIHL patients. Next, we applied the decision tree (J48 in Weka), random forest (RandomForest in Weka), and logistic regression (SimpleLogistic in Weka), supervised machine learning model, to construct gene-based subgroup classification prediction models. All algorithms were trained with a 10-fold cross-validation (CV) in Weka(3.8.6)[24].The random forest also assigns a score to each variant according to the mean decrease in impurity (MDI), which represents the amount of information each variant has for stratification the population according to the genotype of the variant[25]. The importance of variables was ranked using information gain (InfoGainAttributeEval in Weka)[26]. Weka version 3.8.6 default hypyerparameters were used to configure all algorithms.

3.1 Clinical characters baseline

In our hearing test survey, 78 people have a certain frequency > 25 dB, and 35 people have been exposed to noise for 3 years or more, but none of them has hearing loss. We sequenced a total of 78 NIHL patients and 35 control samples by whole exome sequencing. In Table 1, Comparison of age and working time between the two groups have been shown. In Table 2, hearing condition of 78 people with 156 ears in NIHL group and 35 people with 70 ears in control group have been shown. Parameters including age and working hours showed no statistical difference between NIHL group and control group (Table 1). There was significant difference in high frequency average hearing threshold between NIHL group and control group (P < 0.05).

Table 1

Age and work time contrasts between two groups.
	NIHL group (n = 78)	Control group (n = 35)	t	P
Age	34.09 ± 9.68	31.51 ± 6.10	1.712	0.090
Work time (years)	12.15 ± 9.06	9.63 ± 4.99	1.897	0.060

Table 2

Hearing condition between two groups
	NIHL group(n = 156)	Control group(n = 70)	t	P
high frequency threshold	30.44 ± 15.25	15.40 ± 3.17	11.762	< 0.001
speech recognition threshold	17.36 ± 6.98	13.76 ± 3.34	5.234	< 0.001
250 Hz	16.89 ± 7.88	14.00 ± 4.55	3.471	0.001
500 Hz	18.11 ± 7.37	15.29 ± 4.65	3.482	0.001
1 kHz	15.93 ± 7.31	12.36 ± 4.32	4.577	< 0.001
2 kHz	18.53 ± 9.21	13.43 ± 3.67	5.940	< 0.001
3 kHz	23.04 ± 14.86	13.71 ± 4.23	7.218	< 0.001
4 kHz	30.77 ± 18.61	15.50 ± 4.44	9.653	< 0.001
6 kHz	37.50 ± 18.94	17.00 ± 4.85	12.632	< 0.001

3.2 Statistics of SNPs location information and types

After comparing the original sequence with the human genome (hg19), the coverage and average depth were calculated. In most samples, the coverage rate is more than 99%, and the average coverage rate is 99.4%. The average sequence depth range is 20 × to 100 ×, the average depth of the target area of all samples is 44.8.

We selected mutations with mutation frequency < 0.01 in the genome database. Significant difference in allele frequency distribution between cases and controls were observed in one of the 63928 candidate variants. Due to the small number, we also selected the variations detected in at least 5% of the samples in NIHL group but not in any samples in control group. Based on the fact that 73 susceptibility gene exist in > 5% of NIHL cases and do not exist in the control, clustering based on genomic variation spectrum distinguishes two main NIHL subtypes (subtypes A and B) different from the control group. The sensitivity and specificity of unsupervised machine learning grouping were compared among decision tree, random forest model and logistic regression. The AUC was 93.87%. Table 2 shows the performance of the classifier used, and shows that the random forest significantly better than the results of logistic regression and decision tree which show more or less similar results in terms of sensitivity, specificity, accuracy, F-measure and ROC area. These measurements for random forests are: sensitivity = 91.0%, specificity = 91.1%, accuracy = 91.1%, F-measure = 91.0%, and AUC = 97.5%. In terms of AUC, compared with random forest (97.5%), the discrimination ability of decision tree (85.8%) and logistic regression (83.0%) is significantly reduced. After we distinguished subgroup A and subgroup B in NIHL group, we compared the hearing of patients in the two subgroups in Table 4 and the difference was significant in high frequency. The difference between the two subgroups was only significant at 3 kHz (P < 0.05), while not very significant at 4 kHz, 6 kHz and high frequency threshold (P < 0.1), probably because of small sample size or short noise exposure time. It generally showed that the high frequency hearing threshold of subgroup A was worse than that of subgroup B. In the two subgroups determined by susceptibility genes by cluster analysis method, subgroup A has the susceptible genes VWF, COL5A1, COL17A1 and MYO18A. However, subgroup A with these gene mutations had poorer hearing than subgroup B without these genes. Therefore, these gene mutations carried by subgroup A may make hearing worse in noisy environment. The main susceptibility gene contributors of stratification are UTP20, DNAH1, ADAMTS5, SMPD4, COL5A1, SLC12A6, LIPH, MYO18A, CRB2 and SRMS (Importance value was 0.238, 0.225, 0.211, 0.211, 0.204, 0.197, 0.196, 0.193, 0.181 and 0.181 respectively). In cluster analysis and subsequent classification prediction, it can also be concluded that the main susceptibility gene that can mainly distinguish subgroup A is VWF, COL5A1, COL17A1 and MYO18A, and the susceptibility gene that can distinguish subgroup B is UTP20, DNHA1, ADAMTS5, SMPD4. It may be that among the susceptibility genes that distinguish subgroup A from subgroup B, the susceptibility gene carried by subgroup A will make hearing worse, while the gene mutation carried by subgroup B is not very harmful.

Table 3

Performance of the machine learning algorithms.
Model	Sensitivity	Specificity	Accuracy	F-measure	AUC
Logistic regression	85.9%	87.1%	86.7%	86.0%	83.0%
Decision tree	85.9%	87.1%	86.7%	86.0%	85.8%
Random Forest	91.0%	91.1%	91.1%	91.0%	97.5%

Table 4

Subtypes A and B clinical characters
	A subtypes(n = 41)	B subtypes(n = 37)	t	P
Age	34.32 ± 9.07	33.84 ± 10.44	0.217	0.829
Work time (years)	12.24 ± 8.09	12.04 ± 10.13	0.098	0.922
high frequency threshold	32.66 ± 15.81	27.97 ± 14.30	1.935	0.055
speech recognition threshold	17.45 ± 5.48	17.26 ± 8.38	0.166	0.868
250 Hz	16.59 ± 6.52	17.23 ± 9.18	0.500	0.618
500 Hz	18.17 ± 6.11	18.04 ± 8.59	0.110	0.913
1 kHz	15.91 ± 5.94	15.95 ± 8.63	0.026	0.979
2 kHz	19.15 ± 8.60	17.84 ± 9.86	0.885	0.378
3 kHz	25.61 ± 16.13	20.20 ± 12.83	2.301	0.023*
4 kHz	33.48 ± 18.65	27.77 ± 18.23	1.928	0.056
6 kHz	38.90 ± 20.22	35.95 ± 17.41	1.935	0.055
(* P < 0.05)

The global disease burden study measured the number of years of life of people with disabilities [2] and found that hearing loss is the fourth major cause of disability in the world. The main impact of adult hearing loss is communication barriers, which may adversely affect relationships with family and friends and cause difficulties in the workplace. Untreated hearing loss in adults also has indirect health, psychosocial and economic effects, leading to social isolation and decreased quality of life [3, 4, 5].

Prediction models using machine learning can generate robust diagnostic and predictive parameters, because they use the relationship between data to generate accurate predictions, so as to verify their incremental effectiveness[27–29]. Establishing a reasonable machine learning gene analysis model of noise induced hearing loss is not only the basis for exploring the etiology and treatment of noise induced hearing loss on the basis of genetics, but also the basis for noise protection and screening in special posts or specific populations. Mutants of gene VWF appeared in 9 individuals which appears most frequently. Studies have found that VWF is related to NIHL disease in workers exposed to noise[30, 31]. But no more research has been done. Another study showed that VWF may play a role in sudden sensorineural hearing loss in adults and more than half of patients have thrombotic tendencies[32]. These may be related to the pathogenesis of the auditory system. In the future, we can study this gene from the mechanism of hearing loss.

The difference of hearing loss between subgroup A and subgroup B is not significant, and the difference may be mainly in the pathogenesis of hearing loss between the two groups. These people who may have NIHL can only be identified from genomic information. Whether these subtypes have different etiology or other clinical significance needs further study. The internal variability of each individual genome poses a challenge to variation analysis[33]. The presence of > 5% (i.e. 4 of 78 NIHL patients) was considered necessary to identify fixed variants in the NIHL population. Due to our strict screening conditions, the number of possible susceptibility genes in the population is relatively small. We selected 73 genes that are common to at least 4 people and do not exist in 35 people in the control group, and classified and predicted them as features of unsupervised and supervised machine learning.

We identified top ten specific mutations that were predicted to be harmful or destructive, including UTP20, DNAH1, ADAMTS5, SMPD4, COL5A1, SLC12A6, LIPH, MYO18A, CRB2 and SRMS. The mutation of UTP20 is mainly a marker of subgroup B. Possible susceptibility gene VWF, COL5A1, COL17A1, MYO18A exist only in subgroup A but not in subgroup B. And UTP20 was found in 7 people in all NIHL populations and only in subgroup B and Gene Ontology (GO) annotations related to this gene include RNA binding and binding. However, the mechanism of these genes in hearing loss remains to be studied. The above genes play a role in distinguishing the two subgroups. In previous studies, VWF and COL5A1 gene mutations were indeed found to be related to hearing loss[34]. Subgroup A with these genes was indeed worse at high frequency hearing threshold than subgroup B with few of these genes. However, the genes carried by subgroup B that can clearly distinguish subgroup A and subgroup B may be less harmful, and the mechanism needs further study.

The random forest performs better than the decision tree in predicting the classification of subgroups by the possible susceptibility genes we find. This result is expected, because it is known that a random forest with multiple single trees is more robust than a single decision tree. Random forest considers the output of multiple decision trees, which solves their sensitivity to training data, thus reducing the variance in data results.

Although we claim the functional role of the genes involved in the genome variation spectrum, this does not mean that we refer to the causality of NIHL. Not all profile variants need to be directly related to pathology; These variants may be biomarkers, but whether there are differences in the changes of auditory function needs to be studied in a larger sample. Importantly, the subtypes we describe can only be distinguished by their genetic characteristics rather than clinical parameters, suggesting that they can be detected by the molecular depth of genomic analysis so far.

However, this study has some limitations. Firstly, the prediction algorithm used in this study is suitable for a relatively small sample size. Generally speaking, the algorithm used in this study performs better with a larger sample size. Especially for logistic regression models, these algorithms may require 10 times the events of each predictive variable to achieve a small amount of overfitting[35].

We distinguished the genomic variation spectrum between NIHL cases and controls, and also revealed two subtypes of NIHL. Although the genetic and pathogenic nature of variation is unclear, different variation spectra can be used as building blocks of prediction models to detect subtype A in the general population and provide evidence for the use of genomic and personalized medical prediction of noise sensitive procedures. We acknowledge that the good predictive value of our model depends on the 73 susceptibility gene variations correctly predicted.

This is the first study to link NIHL with gene variants and has produced a new potential classification method that can be applied to precision medicine in people’s different sensitivity to noise. We describe the genomic profiles of 73 susceptibility genes associated with NIHL people. The susceptibility map is used to create a prediction model that can identify NIHL individuals and classify them into two genomic subtypes (A and B) with high precision, specificity and sensitivity. We use three machine learning algorithms to predict and classify and finally find that the prediction effect of random forest is the best. Further prospective studies in independent populations are needed to determine the repeatability of the model and evaluate the preventive potential of the two genetic subtypes in clinical practice.

Ethics approval and consent to participate

The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). The study was approved by the ethical committee of PLA Strategic Support Force Characteristic Medical Center (No. K2019(03) ) and informed consent was taken from all individual participants.

Consent for publication

All study subjects signed written informed consent.

Competing interests

The authors have no competing interests to declare.

Availability of data and materials

All data generated and analysed during this study are included in this published article and its additional information files.

Funding

This research received no external funding.

Authors' contributions

Conceptualization, BF, GW, HH and WW; data curation and formal analysis, BL, LW, CQ, XZ and HL; writing-original draft preparation, BF and ZS; writing-review and editing, YZ and YZ. All authors have read and agreed to the published version of the manuscript. All authors read and approved the final manuscript.

Acknowledgements

Not applicable.

Royster JD. Preventing Noise-Induced Hearing Loss. N C Med J. 2017;78(2):113–7.
Feder K, Michaud D, McNamee J, Fitzpatrick E, Davies H, Leroux T. Prevalence of Hazardous Occupational Noise Exposure, Hearing Loss, and Hearing Protection Usage Among a Representative Sample of Working Canadians. J Occup Environ Med. 2017;59(1):92–113.
Moser T, Starr A. Auditory neuropathy–neural and synaptic mechanisms. Nat Rev Neurol. 2016;12(3):135–49.
Willems PJ. Genetic causes of hearing loss. N Engl J Med. 2000;342(15):1101–9.
Marazita ML, Ploughman LM, Rawlings B, Remington E, Arnos KS, Nance WE. Genetic epidemiological studies of early-onset deafness in the U.S. school-age population. Am J Med Genet. 1993;46(5):486–91.
Hong O, Kerr MJ, Poling GL, Dhar S. Understanding and preventing noise-induced hearing loss. Dis Mon. 2013;59(4):110–8.
Konings A, Van Laer L, Van Camp G. Genetic studies on noise-induced hearing loss: a review. Ear Hear. 2009;30(2):151–9.
Sliwinska-Kowalska M, Pawelczyk M. Contribution of genetic factors to noise-induced hearing loss: a human studies review. Mutat Res. 2013;752(1):61–5.
Ohlemiller KK, McFadden SL, Ding DL, Flood DG, Reaume AG, Hoffman EK, et al. Targeted deletion of the cytosolic Cu/Zn-superoxide dismutase gene (Sod1) increases susceptibility to noise-induced hearing loss. Audiol Neurootol. 1999;4(5):237–46.
Ohlemiller KK, McFadden SL, Ding DL, Lear PM, Ho YS. Targeted mutation of the gene for cellular glutathione peroxidase (Gpx1) increases noise-induced hearing loss in mice. J Assoc Res Otolaryngol. 2000;1(3):243–54.
Kozel PJ, Davis RR, Krieg EF, Shull GE, Erway LC. Deficiency in plasma membrane calcium ATPase isoform 2 increases susceptibility to noise-induced hearing loss in mice. Hear Res. 2002;164(1–2):231–9.
Holme RH, Steel KP. Progressive hearing loss and increased susceptibility to noise-induced hearing loss in mice carrying a Cdh23 but not a Myo7a mutation. J Assoc Res Otolaryngol. 2004;5(1):66–79.
Civelek M, Lusis AJ. Systems genetics approaches to understand complex traits. Nat Rev Genet. 2014;15(1):34–48.
Lavinsky J, Crow AL, Pan C, Wang J, Aaron KA, Ho MK, et al. Genome-wide association study identifies nox3 as a critical gene for susceptibility to noise-induced hearing loss. PLoS Genet. 2015;11(4):e1005094.
Liang P, Chen F, Wang S, Li Q, Li W, Wang J, et al. Whole exome sequencing of six Chinese families with hereditary non-syndromic hearing loss. Int J Pediatr Otorhinolaryngol. 2021;148:110817.
Mahfood M, Chouchen J, Kamal Eddine Ahmad Mohamed W, Al Mutery A, Harati R, Tlili A. Whole exome sequencing, in silico and functional studies confirm the association of the GJB2 mutation p.Cys169Tyr with deafness and suggest a role for the TMEM59 gene in the hearing process. Saudi J Biol Sci. 2021;28(8):4421–9.
Ngiam KY, Khor IW. Big data and machine learning algorithms for health-care delivery. Lancet Oncol. 2019;20(5):e262-e73.
Sidey-Gibbons JAM, Sidey-Gibbons CJ. Machine learning in medicine: a practical introduction. BMC Med Res Methodol. 2019;19(1):64.
Deo RC. Machine Learning in Medicine. Circulation. 2015;132(20):1920–30.
DePristo MA, Banks E, Poplin R, Garimella KV, Maguire JR, Hartl C, et al. A framework for variation discovery and genotyping using next-generation DNA sequencing data. Nat Genet. 2011;43(5):491–8.
do Valle IF, Giampieri E, Simonetti G, Padella A, Manfrini M, Ferrari A, et al. Optimized pipeline of MuTect and GATK tools to improve the detection of somatic single nucleotide polymorphisms in whole-exome sequencing data. BMC Bioinformatics. 2016;17(Suppl 12):341.
Wang K, Li M, Hakonarson H. ANNOVAR: functional annotation of genetic variants from high-throughput sequencing data. Nucleic Acids Res. 2010;38(16):e164.
Kosub S. A note on the triangle inequality for the Jaccard distance. Pattern Recogn Lett. 2019;120:36–8.
Witten IH, Witten IH. Data mining: practical machine learning tools and techniques. Fourth Edition. ed. Amsterdam: Elsevier; 2017. xxxii, 621 pages p.
Han H, Guo XL, Yu H. Variable Selection Using Mean Decrease Accuracy And Mean Decrease Gini Based on Random Forest. Int Conf Softw Eng. 2016:219–24.
AlKaabi LA, Ahmed LS, Al Attiyah MF, Abdel-Rahman ME. Predicting hypertension using machine learning: Findings from Qatar Biobank Study. Plos One. 2020;15(10).
Sardaar S, Qi B, Dionne-Laporte A, Rouleau GA, Rabbany R, Trakadis YJ. Machine learning analysis of exome trios to contrast the genomic architecture of autism and schizophrenia. BMC Psychiatry. 2020;20(1):92.
Liu J, Mao R, Ren G, Liu X, Zhang Y, Wang J, et al. Whole Exome Sequencing Identifies Putative Predictors of Recurrent Prostate Cancer with High Accuracy. OMICS. 2019;23(8):380–8.
Zhao Q, Wang F, Chen YX, Chen S, Yao YC, Zeng ZL, et al. Comprehensive profiling of 1015 patients' exomes reveals genomic-clinical associations in colorectal cancer. Nat Commun. 2022;13(1):2342.
Turcot A, Girard SA, Courteau M, Baril J, Larocque R. Noise-induced hearing loss and combined noise and vibration exposure. Occup Med (Lond). 2015;65(3):238–44.
Pettersson H, Burstrom L, Nilsson T. Raynaud's phenomenon among men and women with noise-induced hearing loss in relation to vibration exposure. Noise Health. 2014;16(69):89–94.
Drouet L, Hautefort C, Vitaux H, Kania R, Callebert J, Stepanian A, et al. Plasma Serotonin is Elevated in Adult Patients with Sudden Sensorineural Hearing Loss. Thromb Haemost. 2020;120(9):1291–9.
Nichols RA, Balding DJ. Effects of population structure on DNA fingerprint analysis in forensic science. Heredity (Edinb). 1991;66 (Pt 2):297–302.
Cai Q, Patel M, Coling D, Hu BH. Transcriptional changes in adhesion-related genes are site-specific during noise-induced cochlear pathogenesis. Neurobiol Dis. 2012;45(2):723–32.
van der Ploeg T, Austin PC, Steyerberg EW. Modern modelling techniques are data hungry: a simulation study for predicting dichotomous endpoints. BMC Med Res Methodol. 2014;14:137.

No competing interests reported.

Download PDF

Version 2

posted

You are reading this latest preprint version

Machine Learning Model Identifies Genomic Variation in Noise-Induced Hearing Loss via Whole-exome Sequencing

Status:

Version 2

Abstract

Figures

1. Background

2. Materials And Methods

2.1. Participants and Inclusion Criteria

2.2. Epidemiological survey and hearing test

2.3 Whole Exome Sequencing

2.4 Patient stratification and prediction of noise-induced hearing loss subtypes

3. Results

3.1 Clinical characters baseline

3.2 Statistics of SNPs location information and types

4. Discussion

5. Conclusion

Declarations

References

Additional Declarations

Status:

Version 2