A machine learning approach for estimating Eastern Asian origins from massive screening of Y chromosomal short tandem repeats polymorphisms
- Open Access
- 08.01.2025
- Original Article
Erschienen in:
Abstract
Introduction
East Asia is one of the most critical regions for studying human population evolution and genetic diversity. The genetic composition of contemporary East Asian populations reflects a complex history shaped by human colonization, differentiation, migrations, and admixture. Over the past few decades, substantial genetic data on East Asian populations have been accumulated [1], fostering efforts to identify and utilize informative markers that can further delineate their genetic substructure [2‐4]. Earlier phylogenetic studies have revealed a north-south pattern of genetic differentiation among East Asian populations, approximately divided by the Yangtze River [5‐9]. Unfortunately, most current forensic panels for ancestry inference classify East Asian populations as a single cluster, thereby limiting their efficacy in precisely determining the specific origin of unknown individuals within East Asia.
To trace the ancestry of East Asians, marked by complex genetic structures, the utilization of effective ancestry informative markers (AIMs) is essential. AIMs, which exhibit substantial allele frequency differences between geographically distinct populations, have been employed to reveal genetic differentiation among human groups across multiple continents. In particular, Y chromosomal genetic markers are favored for initial ancestry inference due to their exclusive paternal transmission and absence of recombination, which preserves distinct population-specific haplotypes [10]. Although Y-STR haplotype data have been extensively utilized to investigate the differentiation and migration patterns of East Asian populations, their application as a tool for inferring the ancestral origins of individuals within this region remains limited.
Anzeige
Meanwhile, to discover the genetic structure of populations, researchers have employed dimensionality reduction techniques such as principal component analysis (PCA) and have examined the proportions of genetic components of studied individuals using software like STRUCTURE and ADMIXTURE [4]. Nonetheless, these conventional approaches have limitations in accurately inferring ancestral origins for populations with closely related genetic backgrounds. Machine learning (ML) has emerged as a promising alternative methodology, enabling computational systems to autonomously learn and uncover complex attributes from high-dimensional datasets. A notable strength of ML lies in its proficiency in uncovering hidden genetic structures that may be overlooked by traditional techniques. With its robust performance in pattern recognition and classification of samples, ML offers significant potential for enhancing the efficacy of tools used for ancestry inference. Indeed, recent studies in the field of population genetics have demonstrated the remarkable utility of ML not only in analyzing genetic structure but also as advanced tool for ancestry prediction [11‐14].
In this study, we assembled a comprehensive dataset of Y-STR haplotypes from Asian populations to characterize genetic differentiation between East Asian groups and their neighboring regions. We utilized various ML techniques, including Support Vector Machine (SVM), XGBoost, and Random Forest classifiers, to develop a classification system among Asian groups. To date, large-scale screening of Y-STR haplotypes utilizing ML approaches has not yet been reported in Asian populations. This study underscores the potential of using advanced computational methods to analyze complex genetic data and reveal population-specific patterns, thereby improving our understanding of genetic diversity and the resolution of ancestry inference.
Materials and methods
Collection of Y-STR profiles
A total of 10,154 unrelated Asian male Y-STR haplotypes were analyzed in this study. Of these, 1,008 DNA samples were either purchased from the Coriell Institute for Medical Research (Camden, NJ, USA) or directly collected as blood, blood FTA cards, or buccal swabs after obtaining written informed consent from all participants (Supplementary Table S1 [15]). The collection of samples for this study received approval from the Institutional Review Board of Seoul National University Hospital Biomedical Research Institute (IRB No. 1404-068-572). DNA extraction was carried out using the QIAamp® DNA Mini Kit (QIAGEN, Hilden, Germany) for blood samples and the QIAamp® DNA Investigator Kit (QIAGEN) for buccal swabs and FTA cards. Quantification of DNA was performed using the Qubit™ Flex Fluorometer (Invitrogen™, MA, USA) with the Qubit™ 1X dsDNA High Sensitivity (HS) Assay Kit (Invitrogen™) following the manufacturer’s guidelines. Multiplex amplification of Y-STR markers was executed using the PowerPlex® Y23 System (Promega, Madison, WI, USA). The amplified products were separated on the 3500 Genetic Analyzer (Applied Biosystems™, Foster City, CA, USA), and the electrophoresis data were analyzed using the GeneMapper® ID Software v.1.5 (Applied Biosystems™). Microvariants, multi-allele, and null alleles were verified through repeated amplification procedures. The remaining 9,146 Y-STR profiles were gathered from various literature sources (Supplementary Table S2 [16‐35]).
For this study, 17 Y-STR loci from the AmpFLSTR™ Yfiler™ PCR Amplification Kit (Applied Biosystems™) were utilized, including DYS438, DYS393, DYS385a/b, DYS389I/II, DYS458, DYS437, DYS391, DYS392, DYS635, Y GATA H4, DYS19, DYS390, DYS439, DYS456, and DYS448, as these were the most abundant markers available within the entire dataset. The Asian populations were categorized into three geographical groups: Northeast Asian (NEA), covering Korea, Northern China, Japan and Mongolia; Southeast Asian (SEA), covering Southern China, Taiwan, Malaysia, the Philippines, Myanmar, and Vietnam; and Southwest Asian (SWA), covering India, Bangladesh, Nepal, Pakistan, and Sri Lanka. The Yangtze River in China served as a geographic boundary to distinguish between NEA and SEA populations, a method commonly employed in other studies.
Anzeige
Population study
Allele frequencies of Y-STR loci and haplotype frequencies were determined by direct counting. Allelic diversity and haplotype diversity (HD) were calculated using Nei’s formula, (n/n − 1) (1 − Σpi2) [36]. The haplotype match probability (HMP) was calculated using the formula, HMP = Σpi2, where pi represents the frequency of the haplotype. The discrimination capacity (DC) was estimated by assessing the proportion of unique haplotypes in each dataset. Population pairwise genetic distances (Rst values) and corresponding P-values were calculated for all given pairs among 15 subgroups (Northern Chinese, Japanese, Korean, Chinese Mongolian, Southern Chinese, Taiwanese, Malaysian, Filipino, Burmese, Vietnamese, Indian, Bangladeshi, Nepalese, Pakistani, and Sri Lankan) in each geographical group using Arlequin version 3.5.2.2 [37]. The analysis of molecular variance (AMOVA) was also performed by the same software to assess the statistical significance of genetic differentiation among groups [38]. The DYS389I PCR product represents a subset of the DYS389II amplicon, as the forward PCR primer binds to the flanking region of two repeat regions that are approximately 120 bp apart. To independently assess the variation in these regions, the number of repeats at DYS389II was calculated by subtracting the DYS389I value [39]. The DYS385 locus was excluded from the distance calculations due to the indistinguishability between its two alleles (DYS385a and DYS385b), and any samples demonstrating non-consensus alleles or null alleles at any locus were also excluded from these analyses. To illustrate the relationships between populations based on the calculated Rst values, a multidimensional scaling (MDS) plot was generated using IBM SPSS Statistics version 29, and a Neighbor-Joining (NJ) tree was constructed using MEGA software version 11.0.13 [40].
Machine learning based ancestry inference
Data preprocessing and development of machine learning models were conducted using the Python programming language. Prior to model construction, the DYS385 locus and haplotypes containing non-consensus or null alleles were excluded, as these elements were not representative of the broader population distribution and could introduce potential bias. The entire dataset was randomly split into a training set (80%) and a test set (20%).
Supervised machine learning was performed using three algorithms from the scikit-learn library [41]: Support Vector Machine (SVM), XGBoost, and Random Forest classifiers. SVM, widely used for classification tasks, aims to identify the optimal hyperplane that maximally separates different classes based on kernel functions. To address the multi-class classification problem, one-against-all (OAA, also known as one-vs-rest) and one-against-one (OAO, also known as one-vs-one) strategies were employed (see Supplementary Fig. S1). To minimize data distortion from outliers, scaling was performed using the RobustScaler from scikit-learn. Additionally, two decision tree-based ensemble algorithms, XGBoost and Random Forest, were used. XGBoost constructs trees sequentially, with each tree trained to correct errors made by previous trees. In contrast, Random Forest builds multiple independent decision trees and combines their predictions (see Supplementary Fig. S2). Each classifier was trained on the training set with hyperparameter tuning and 5-fold cross-validation using grid search to mitigate overfitting and underfitting. Final evaluations were conducted on the test dataset. To ensure robust reproducibility, each experiment was repeated five times.
Model performance was evaluated using a confusion matrix, which categorizes predictions into four outcomes: true positive (TP), indicating the number of correctly predicted positive samples; true negative (TN), indicating the number of correctly predicted negative samples; false positive (FP), representing negative samples incorrectly classified as positive; and false negative (FN), representing positive samples incorrectly classified as negative. From these outcomes, accuracy, precision, recall, and F1 score were calculated using the ‘classification_report’ function. Given the class imbalance in the dataset, both macro and weighted averages were assessed. The macro average calculates metrics by treating all classes equally, whereas the weighted average applies weights based on the size of each class. The equations for these metrics are as follows:
$$\:\text{A}\text{c}\text{c}\text{u}\text{r}\text{a}\text{c}\text{y}=\:\frac{TP+TN}{TP+TN+FP+FN}$$
(1)
$$\:\text{P}\text{r}\text{e}\text{c}\text{i}\text{s}\text{i}\text{o}\text{n}=\:\frac{TP}{TP+FP}$$
(2)
$$\:\text{R}\text{e}\text{c}\text{a}\text{l}\text{l}=\:\frac{TP}{TP+FN}$$
(3)
$$\:\text{F}1\:\text{S}\text{c}\text{o}\text{r}\text{e}=\:\frac{TP}{TP+0.5\:(FP+FN)}$$
(4)
Results
Genetic diversity and population structure of Y-STRs in Asian populations
The allele frequencies and allelic diversities of 15 single-copy Y-STR loci and one multi-copy Y-STR locus are presented across each geographical group in Supplementary Table S3, Table S4, and Fig. S3. The multi-copy Y-STR locus, DYS385, demonstrated the highest level of diversity across all groups (NEA, SEA, and SWA). Among the single-copy loci, DYS458 emerged as the most diverse marker in the NEA and SEA groups, whereas DYS635 displayed the greatest variability in the SWA group. DYS391 and DYS437 exhibited lower levels of polymorphism in all groups, with DYS438 showing the least allelic diversity value particularly in the SEA group. The discrimination capacities for the NEA, SEA, and SWA populations were 0.8726, 0.8612, and 0.8506, respectively. Haplotype analysis identified a high proportion of unique haplotypes (approximately 90%) within each group, as detailed in Supplementary Table S5. A total of 30 haplotypes were common between the NEA and SEA groups, with 6 haplotypes shared between the NEA and SWA groups, and another 6 haplotypes common between the SEA and SWA populations. Notably, two haplotypes were found to be identical across all three groups.
AMOVA results revealed that the highest level of variation exists within 15 subgroups (90.30%), followed by variation among the three geographical groups (5.91%) and among subgroups within geographical groups (3.78%) (Supplementary Table S6). Pairwise genetic distances (Rst values) disclosed a close relationship between the NEA and SEA groups (Rst = 0.0365), while the distance between SEA and SWA was the greatest among the pairs (Rst = 0.1089) (Supplementary Table S7 and Table S8). The MDS plot and NJ tree (Supplementary Fig. S4 and Fig. S5), based on pairwise Rst values, demonstrated significant genetic differentiation between East Asian populations and the SWA group. Remarkably, SEA populations formed a distinct cluster, except for the Burmese population, which showed a pronounced genetic similarity with the Northern Chinese population (Rst = 0.0043).
Machine learning-based ancestry inference among three Asian groups
To differentiate East Asian group and its neighboring populations, four machine learning models were developed and evaluated to infer the geographical origins of individuals based on their Y-STR profiles. Results from five experimental repetitions are presented in Table 1 as the mean ± standard deviation. The optimized hyperparameters for the best-performing model in each classifier are listed in Table 2, with the corresponding confusion matrices shown in Fig. 1.
Table 1
Evaluation metrics for classification models of Asian geographical groups
Model | Accuracy (%) | Precision | Recall | F1 score |
|---|---|---|---|---|
OAO-SVM | 80.73 ± 0.00 | 0.81 ± 0.00 (0.81 ± 0.00) | 0.79 ± 0.00 (0.81 ± 0.00) | 0.80 ± 0.00 (0.81 ± 0.00) |
OAA-SVM | 80.48 ± 0.00 | 0.80 ± 0.00 (0.80 ± 0.00) | 0.79 ± 0.00 (0.80 ± 0.00) | 0.79 ± 0.00 (0.80 ± 0.00) |
XGBoost | 81.82 ± 0.00 | 0.81 ± 0.00 (0.82 ± 0.00) | 0.81 ± 0.00 (0.82 ± 0.00) | 0.81 ± 0.00 (0.82 ± 0.00) |
RandomForest | 82.92 ± 0.08 | 0.83 ± 0.00 (0.83 ± 0.00) | 0.81 ± 0.00 (0.83 ± 0.00) | 0.82 ± 0.00 (0.83 ± 0.00) |
Anzeige
Table 2
Suitable hyperparameters for classification models of Asian geographical groups
Model | Hyperparameter |
|---|---|
OAO-SVM | ‘estimator__C’: 10, ‘estimator__gamma’: ‘scale’ |
OAA-SVM | ‘estimator__C’: 10, ‘estimator__gamma’: ‘scale’ |
XGBoost | ‘learning_rate’: 0.1, ‘max_depth’: 5, ‘n_estimators’: 500 |
RandomForest | ‘max_depth’: 15, ‘min_samples_leaf’: 2, ‘min_samples_split’: 5, ‘n_estimators’: 300 |
Fig. 1
Confusion matrices of the best-performing models for Asian classification: (a) OAO-SVM, (b) OAA-SVM, (c) XGBoost, and (d) Random Forest. * Darker shades represent higher counts
All models demonstrated generalization without overfitting or underfitting. Based on test accuracy, the random forest model achieved the highest performance (82.92%), followed by XGBoost (81.82%), OAO- SVM (80.73%), and OAA- SVM (80.48%). Performance across precision, recall, and F1 scores was consistent among the models, with values ranging from 0.79 to 0.83. For predictive performance across geographic groups, the SEA group consistently showed lower F1 scores than the other groups (Fig. 2). According to the confusion matrix, the NEA and SWA groups achieved prediction accuracies above 80% in all models, whereas accuracy for the SEA group ranged between 66.21% and 70.31% (Fig. 1).
Fig. 2
F1-score of each geographical group in Asian classification models
Anzeige
To further analyze the factors contributing to the high performance of the random forest model, feature importance was extracted and presented in Fig. 3. DYS438, DYS392, and DYS635 exhibited particularly high importance in the model, while DYS391 and DYS437 showed the lowest importance.
Fig. 3
Feature importance of the best-performing random forest models for Asian classification
Machine learning-based ancestry inference between two east Asian groups
Refined models were developed and evaluated specifically for distinguishing between two East Asian groups. Results from five experimental repetitions are presented in Table 3 as the mean ± standard deviation. The optimized hyperparameters for the best-performing model in each classifier are listed in Table 4 with the corresponding confusion matrices shown in Fig. 4.
Table 3
Evaluation metrics for classification models of east Asian geographical groups
Model | Accuracy (%) | Precision | Recall | F1 score |
|---|---|---|---|---|
SVM | 83.31 ± 0.00 | 0.83 ± 0.00 (0.83 ± 0.00) | 0.80 ± 0.00 (0.83 ± 0.00) | 0.81 ± 0.00 (0.83 ± 0.00) |
XGBoost | 83.88 ± 0.00 | 0.83 ± 0.00 (0.84 ± 0.00) | 0.82 ± 0.00 (0.84 ± 0.00) | 0.82 ± 0.00 (0.84 ± 0.00) |
RandomForest | 84.98 ± 0.16 | 0.85 ± 0.00 (0.85 ± 0.00) | 0.82 ± 0.00 (0.85 ± 0.00) | 0.83 ± 0.00 (0.85 ± 0.00) |
Anzeige
Table 4
Suitable hyperparameters for classification models of east Asian geographical groups
Model | Hyperparameter |
|---|---|
SVM | ‘C’: 10, ‘gamma’: 0.1 |
XGBoost | ‘learning_rate’: 0.1, ‘max_depth’: 5, ‘n_estimators’: 500 |
RandomForest | ‘max_depth’: 15, ‘min_samples_leaf’: 2, ‘min_samples_split’: 5, ‘n_estimators’: 200 |
Fig. 4
Confusion matrices of the best-performing models for East Asian classification: (a) SVM, (b) XGBoost, and (c) Random Forest. * Darker shades represent higher counts
All models demonstrated generalization without overfitting or underfitting. Based on test accuracy, the random forest achieved the highest performance (84.98%), followed by XGBoost (83.88%), and SVM (83.31%). Performance across precision, recall, and F1 scores was consistent among the models, with values ranging from 0.80 to 0.85. In terms of predictive performance across geographic groups, the SEA group consistently showed lower F1 scores than the NEA group (Fig. 5). According to the confusion matrix, the NEA group was predicted with 90% accuracy, whereas the SEA group achieved only 70% accuracy, with approximately 365 of 512 test samples correctly classified (Fig. 4).
Fig. 5
F1-score of each geographical group in East Asian classification models
To further analyze the factors contributing to the high performance of the random forest model, feature importance was extracted and presented in Fig. 6. DYS392, DYS438, and DYS390 exhibited particularly high importance in the model, while DYS391 and DYS437 showed the lowest importance.
Fig. 6
Feature importance of the best-performing random forest models for East Asian classification
Discussion
Asia, home to 60% of the world’s population is characterized by diverse ethnic groups with complex genetic structures. Population genetic studies of Asian populations have advanced significantly since the initial Human Genome Organization (HUGO) Pan-Asian SNP Consortium [42] and now include large-scale efforts such as the GenomeAsia 100 K project [43]. The accumulation of genetic data and advancements in experimental techniques have made it possible to study both ancient and modern human genomes with high precision. This progress has enabled researchers to investigate the genetic history of Asian populations over the past 45,000 years, revealing differentiation into three primary ancestral lineages: Ancient Ancestral South Indian (AASI), Australasian (AA), and East and Southeast Asian (ESEA) [44]. Consistent with prior studies, the genetic structure analysis conducted in this study demonstrates distinct genetic differentiation between Southwest Asian and East Asian groups. Furthermore, our extensive Y-STR analysis supports the notion that East Asian populations can be genetically subdivided into northern and southern clusters along the Yangtze River. Within the East Asian cluster, subgroup analysis revealed a close genetic relationship between Korean and Japanese populations, which likely originates from the geographical connectivity between the Korean Peninsula and Japanese archipelago before their separation due to rising sea levels [45]. Meanwhile, the Northern Chinese subgroup exhibits high genetic affinity with Southeast Asian groups, particularly the Myanmar population. This pattern may be attributed to historical migration patterns, such as the southward movement of Tibeto-Burman populations from northwestern China approximatel 2,600 years ago [46]. However, when inferring specific ethnic relationships, it is important to note that historical factors, including social structures and marriage patterns, may impact interpretation. Furthermore, results can vary significantly depending on the reference datasets used.
To gain a deeper understanding of the complex genetic structure of continental regions, researchers have increasingly employed machine learning approaches in biogeographical ancestry studies. However, for East Asians, who exhibit intricate migration patterns, research remains limited. Sun et al. [4] developed a method to predict the geographic origins of East Asian populations using insertion-deletion polymorphisms(InDels) and machine learning algorithms. Their study demonstrated the outstanding performance of ML methods for forensic ancestry inference, achieving an accuracy of approximately 80% using multi-indel markers, the best performing model achieving 87%. It is important to consider that the cohort used in their study differed from the subjects in our research. The authors included Malaysian, Laotian, and Cambodian samples in the Southeast Asian group but excluded ethnic groups adjacent to Northeast Asia, such as the Southern Chinese populations. The individual genotypic differences in their training data may have contributed to more effective learning and classification compared to the groups we studied. In contrast, given the extensive diversity among Asian populations, our model aimed to predict geographic origin by incorporating data from broad regions. To improve classification of East Asian populations, this study developed a model to identify East Asians among broader Asian groups and further categorize them into Northeast and Southeast clusters. Machine learning approaches, with their ability to efficiently handle large-scale datasets, were particularly well-suited for this task and offer flexibility for future data integration.
In this study, we selected three classification algorithms. SVM was chosen for its effectiveness in handling high-dimensional data, while decision tree models were utilized for their ability to provide interpretable predictions through tree-based structures. Our findings indicated that tree-based models using raw data outperformed SVM models with transformed data. Although it is challenging to pinpoint the exact reasons for this performance difference due to SVM’s black-box nature, the superior performance of tree-based models may be attributed to their capacity for capturing potential interactions between markers, or alternatively, the scaling applied to SVM might have diluted the relationships between STR values. In forensic DNA analysis, the transparency of the result-generation process is crucial [47]. The combination of superior performance and the white-box nature of tree-based models, which allows for visual examination of the model construction process, enhances transparency and reliability in forensic applications.
Four key metrics were investigated to evaluate model performance. Accuracy represents the proportion of correctly predicted samples in the entire sample set. Precision indicates the ratio of true positives among all samples predicted as positive, making it particularly useful when minimizing false positives is essential. Recall represents the ratio of true positives among all actual positive samples, which is crucial when avoiding false negatives. The F1 score, calculated as the harmonic mean of precision and recall, is widely used in evaluating imbalanced datasets in multi-class classification [48]. Given the slight class imbalance in this study, we focused on both accuracy and F1 score. All machine learning models demonstrated the capability to classify Asian geographical groups with accuracy exceeding 80% and achieved F1 scores of approximately 0.80 overall (Tables 1 and 3). High F1 scores generally indicate balanced performance, suggesting that the models achieved both high precision and high recall. In this study, models maintained stable performance with F1 scores consistently around 0.80. However, in subgroup-specific analyses, the SEA group showed relatively lower F1 scores (Figs. 2 and 5). This performance difference likely results from the smaller training dataset available for this group compared to others. Future research should incorporate additional data for Southeast Asian populations.
Compared to classical statistical analysis, machine learning approaches have the advantage of analyzing patterns among all genetic markers in the training dataset, selecting those with the highest predictive power. Feature importance indicates the contribution of each feature to tree-based decision-making, measured on a scale from 0 to 1. Values closer to 1 indicate greater predictive accuracy, and the sum of all feature importance values equals 1 [48]. Markers with low mutation rates are generally suitable for ancestry inference, as they better reflect human history. While stable markers were expected to have high importance in machine learning models, our exploratory analysis did not reveal statistically significant relationships between feature importance and genetic characteristics such as mutation rates or genetic diversity. This suggests that the model’s feature selection may be influenced by complex patterns beyond basic genetic attributes, although further investigation with an expanded set of markers would be necessary to confirm this observation. Future research could benefit from employing a broader array of STR markers, selecting only those that are most informative, potentially leading to the development of more effective models. However, it is essential to recognize that feature importance rankings are specific to the model used, necessitating the careful selection of markers that align with the particular model employed.
Meanwhile, the Y chromosome serves as a potent genetic marker due to its capacity to provide male-specific genetic information, which is particularly valuable in forensic cases involving sexual assaults where distinguishing male DNA from a female background is crucial [49]. This study selected Y-chromosomal markers to utilize these forensic advantages. Y-STR markers are particularly significant, as they are widely represented in public databases, owing to their integration into standard genetic testing procedures. Previous research has shown that Y-chromosomal single nucleotide polymorphisms (Y-SNPs) possess strong geographical specificity, enabling direct assessment of admixture patterns among diverse populations [50, 51]. Additionally, Y-chromosomal insertion-deletion polymorphisms (Y-InDels) are useful tools for tracing biogeographical origins [52]. The strategic combination of these Y-chromosomal markers is expected to enhance the resolution of population genetic analyses [53]. It is important to note, however, that DNA variants on the Y chromosome represent a single paternal lineage and may not fully capture an individual’s overall ancestry, particularly when distant maternal lineages with unusual ancestry are inherited [10]. This limitation can be addressed by incorporating autosomal genetic markers. Future studies that include diverse marker types may provide a more comprehensive understanding of population genetic history.
Conclusion
In this study, we conducted a comprehensive analysis of Y-STR haplotypes across a broad dataset of Asian populations to explore their genetic diversity and differentiation. Our analysis highlighted distinct genetic profiles between East Asian and Southwest Asian groups, underscored by a pronounced north-south pattern of genetic differentiation within East Asia. The deployment of machine learning techniques showed high accuracy in classifying individuals into their respective Asian groups based on Y-STR haplotypes. These findings offer valuable insights into the ancestry of populations in East Asia and adjacent regions, presenting machine learning models based on Y-STR markers as promising predictive tools. Future research, incorporating a broader array of genetic markers and enhanced computational techniques, could further improve the resolution of ancestry.
Acknowledgements
The authors express sincere gratitude to all the volunteers who donated DNA samples for this study. We would like to express our appreciation to Prof. Dr. Vasant Shinde of Deccan College for kindly providing the samples. Special thanks are extended to Mr. Yoon Geon Kim for his invaluable contribution in providing the initial machine learning model.
Declarations
Ethical approval
Approval was obtained from the Institutional Review Board of Seoul National University Hospital Biomedical Research Institute (IRB No. 1404-068-572). The procedures used in this study adhere to the tenets of the Declaration of Helsinki.
Consent to participate
Informed consent was obtained from all individual participants included in the study.
Competing interests
The authors have no competing interests to declare that are relevant to the content of this article.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
