Next Article in Journal
A Pilot Study of Klebsiella pneumoniae in Community-Acquired Pneumonia: Comparative Insights from Culture and Targeted Next-Generation Sequencing
Previous Article in Journal
Endothelial Activation and Stress Index (EASIX) Predicts In-Hospital Mortality in Acute Decompensated Heart Failure with Reduced Ejection Fraction
Previous Article in Special Issue
Association of Hemoglobin to Red Blood Cell Distribution Width Ratio and Total Bone Mineral Density in U.S. Adolescents: The NHANES 2011–2018
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Diagnostics
Volume 16
Issue 1
10.3390/diagnostics16010153
diagnostics-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Integrated Cross-Platform Analysis Reveals Candidate Variants and Linkage Disequilibrium-Defined Loci Associated with Osteoporosis in Korean Postmenopausal Women

by
Su Kang Kim
1,
Seoung-Jin Hong
2,
Seung Il Song
3,
Jeong Keun Lee
3,
Gyutae Kim
4,
Byung-Joon Choi
5,
Suyun Seon
6,
Seung Jun Kim
6,
Ju Yeon Ban
7 and
Sang Wook Kang
8,*
1
Department of Biomedical Laboratory Science, Catholic Kwandong University, Gangneung 25601, Republic of Korea
2
Department of Prosthodontics, College of Dentistry, Kyung Hee University, Seoul 02447, Republic of Korea
3
Department of Dentistry, School of Medicine, Ajou University, Suwon 16499, Republic of Korea
4
Department of Oral and Maxillofacial Radiology, College of Dentistry, Kyung Hee University, Seoul 02447, Republic of Korea
5
Department of Oral and Maxillofacial Surgery, College of Dentistry, Kyung Hee University, Seoul 02447, Republic of Korea
6
Department of Dentistry, Graduate School, Kyung Hee University, Seoul 02447, Republic of Korea
7
Department of Dental Pharmacology, College of Dentistry, Dankook University, Cheonan 31116, Republic of Korea
8
Department of Oral and Maxillofacial Pathology, College of Dentistry, Kyung Hee University, Seoul 02447, Republic of Korea
*
Author to whom correspondence should be addressed.
Diagnostics 2026, 16(1), 153; https://doi.org/10.3390/diagnostics16010153
Submission received: 30 November 2025 / Revised: 23 December 2025 / Accepted: 30 December 2025 / Published: 3 January 2026
(This article belongs to the Special Issue Current Diagnosis and Management of Metabolic Bone Disease)
Browse Figures
Versions Notes

Abstract

Background: Osteoporosis is highly prevalent in postmenopausal women, yet genome-wide association studies often miss disease-relevant variants because of incomplete single nucleotide polymorphism (SNP) coverage and platform-specific limitations. We aimed to identify genetic contributors to osteoporosis risk by integrating two exome-based genotyping platforms with multilayer analytic approaches. Methods: We analyzed extreme osteoporosis phenotypes in Korean postmenopausal women from the Korean Genome and Epidemiology Study (KoGES) Ansan–Anseong cohorts using the Illumina Infinium HumanExome BeadChip and the Affymetrix Axiom Exome Array. After standard quality control, single-SNP logistic regression, cross-platform overlap analysis, and three machine-learning models were applied. Predicted functional impact was evaluated using multiple in silico algorithms and conservation scores. Finally, datasets from both platforms were merged, and cross-platform linkage disequilibrium (LD) blocks were defined to identify loci containing SNPs with p < 1 × 10−4. Results: No overlapped SNP reached genome-wide significance, but rs2076212 in PNPLA3 achieved suggestive significance (p < 1 × 10−5) only on the Illumina array. Cross-platform analysis identified 111 overlapping SNPs in 70 genes. Integrated machine-learning, in silico, and conservation evidence prioritized ARMS2, CCDC92, NQO1, ZNF510, PTPRB, and DYNC2H1 as candidate genes. LD-block analysis revealed 10 blocks with at least one SNP at p < 1 × 10−4, including four chromosome 12 loci (NAV2, BICD1, CCDC92, ZNF664) that became apparent only when LD patterns were evaluated jointly across platforms. Conclusions: Combining dual exome arrays with LD-block analysis, machine learning, and functional prediction improved sensitivity for detecting low bone mineral density-related loci and highlighted CCDC92, DYNC2H1, NQO1, and related genes as biologically plausible candidates for future validation.

1. Introduction

Osteoporosis is a disease characterized by weakened bones that become easily fractured due to low bone mineral density (BMD) and microstructural changes. It is particularly prevalent among postmenopausal women and the older adults. The World Health Organization (WHO) defines osteoporosis as a condition where bone mineral density is −2.5 or lower when comparing bone density measurements to the average bone density of young adults [1]. Osteoporosis is a widespread health issue globally, affecting over 200 million people aged 50 and older. According to the Korea Disease Control and Prevention Agency, 37.3% of women and 7.5% of men over the age of 50 in South Korea have been diagnosed with osteoporosis (www.kdca.go.kr). The prevalence of osteoporosis increases rapidly with age, affecting approximately two-thirds of women and one-fifth of men aged 70 and older.
Despite being a common condition, osteoporosis can have severe, even fatal, consequences. The most significant risk associated with osteoporosis is fractures, which can lead to serious complications. Hip fractures, in particular, often result in mobility loss and prolonged bed rest, increasing the risk of complications such as bedsores and pneumonia. With a mortality rate of approximately 20% within one year following a hip fracture, osteoporosis should therefore be regarded as a serious health threat [2].
The development of osteoporosis is influenced by a wide range of factors, including age, menopause, lifestyle habits such as smoking, alcohol consumption, and physical inactivity, as well as nutritional deficiencies and the use of certain medications [3]. Among these risk factors, genetic predisposition has been increasingly recognized as a major contributor. Numerous studies have shown that genetic influences substantially affect both the onset of osteoporosis and the reduction in bone mineral density. Twin studies, in particular, have indicated that up to 60–80% of the variance in bone density can be attributed to genetic factors, underscoring the importance of genetic predisposition in the pathogenesis of osteoporosis risk or low BMD [4]. To date, most genetic research on osteoporosis risk or low BMD has been conducted using genome-wide association studies (GWAS), which are widely used to identify genetic variants associated with complex diseases [5]. Previous GWAS investigating bone mineral density (BMD) in Korean populations have mainly focused on DXA-based phenotypes and used genotyping or multi-ethnic meta-analyses to improve genome coverage, but have typically relied on single genotyping platforms [6,7]. However, incomplete SNP coverage, reduced power to detect rare variants, and the limited coverage of exonic variants and the dependence on a single platform can lead to missed trait-relevant loci [8].
To address these limitations, this study employed two complementary genotyping platforms: the Illumina Exome Chip and the Affymetrix Axiom Exome Array. Using both arrays expands SNP coverage through distinct probe sets, enhances the detection of rare variants, enables cross-validation of significant loci to reduce false positives and negatives, and mitigates platform-specific biases, thereby improving the reliability and reproducibility of GWAS results. The Illumina Exome Chip and the Affymetrix Axiom Exome Array, commonly used platforms for genetic analysis, each possess distinct advantages and limitations. The Illumina Exome Chip is tailored to rare coding variants such as missense and nonsense mutations, but its design—based primarily on European-ancestry exomes—limits representation in non-European populations, and array-based genotyping constraints result in ~20–30% of targeted variants failing probe design or accurate genotyping. Moreover, statistical power for detecting low-frequency and especially ultra-rare variants remains limited [9,10,11]. In contrast, the Affymetrix Axiom Exome Array exhibits high accuracy and reproducibility across diverse variants, yet shows a marked decline in positive predictive value for ultra-rare heterozygous calls, leading to false positives [12,13].
While analyzing two genotyping platforms simultaneously is highly valuable, single-SNP association results can appear inconsistent—even when reflecting the same underlying biological signal—due to differences in variant density, probe design, and tagging efficiency. To address these limitations, this study incorporated LD block–level analysis to group correlated SNPs into coherent genomic loci [14,15]. This approach enables key SNPs from both platforms to be integrated within the same LD-defined regions, thereby improving the robustness of cross-platform validation, particularly for rare or poorly tagged variants.
Therefore, the aim of this study is to simultaneously analyze data obtained from the Illumina Infinium HumanExome BeadChip and the Affymetrix Axiom Exome Array and to cross-validate the findings from each platform in order to more accurately identify genes associated with low BMD and osteoporosis risk. The ultimate goal is to elucidate the genetic underpinnings of low BMD and osteoporosis risk, thereby providing foundational insights that can inform future efforts in disease prediction, diagnosis, and the development of personalized treatment strategies.

2. Materials and Methods

2.1. Study Subjects

This study was conducted using data from the third phase of the Anseong and Ansan cohorts of the Korean Genome and Epidemiology Study (KoGES), which included a total of 7077 subjects. Figure 1 presents a flowchart outlining the subject selection and analysis process. Initially, men and premenopausal women were excluded, and only postmenopausal women were included in the analysis. In this study, we used bone status indices based on Sunlight OmniSense quantitative ultrasound (QUS) measurements to define osteoporosis risk. Previous studies have reported that QUS measurements are significantly correlated with DXA measurements, particularly in postmenopausal women [16]. T- and Z-scores were derived from speed-of-sound (SOS) measurements [17]. Specifically, low BMD was defined as participants with a T-score < −2.5 and a Z-score < −2.0, as measured by QUS at the midshaft tibia or distal radius. This operational definition was used to maximize phenotypic contrast and differs from the WHO/ISCD clinical diagnostic criteria [1], which rely on central DXA measurements at the lumbar spine or hip. Healthy controls were defined as participants with a QUS-derived T-score ≥ −1.0, and to further enhance separation between cases and controls, participants with T-scores between −2.5 and −1.0 were excluded from the analysis. To isolate the genetic influence on low BMD, subjects with known risk factors such as body weight < 58 kg, body mass index (BMI) < 19, alcohol consumption > 10 mL/day, smoking, long-term steroid use, hormone therapy, and a medical history of fracture or arthritis were also excluded [1,18].
The final selected participants were as follows:
•
Illumina Infinium HumanExome BeadChip (Illumina, Inc., San Diego, CA, USA) group: 98 healthy controls and 191 low BMD cases (total n = 289).
•
Affymetrix Axiom Exome Array group: 99 healthy controls and 194 low BMD cases (total n = 293).
From the Illumina Infinium HumanExome BeadChip, 30,538 single nucleotide polymorphisms were analyzed, and a total of 1200 SNPs met the quality control (QC) thresholds of p < 0.05, minor allele frequency (MAF) > 0.05, and Hardy–Weinberg equilibrium (HWE) > 0.01, which were applied to minimize potential false-positive associations [19]. From the Affymetrix Axiom Exome Array (Thermo Fisher Scientific Inc., Waltham, MA, USA), 242,901 SNPs were examined, and 15,703 SNPs met the same criteria. A cross-platform comparison revealed that 111 SNPs across 70 genes were commonly identified as statistically significant in both datasets. This study was approved by the Institutional Review Board of Dankook University (IRB No. 2018-08-004 Date: 10 September 2018). As indicated in the IRB documents, this study was classified as a retrospective cohort/genetic data–based analysis, and therefore written informed consent was waived. The IRB approved this exemption and included a “Written Consent Waiver Statement” in the official documentation. Accordingly, no participant consent form was required, and none was collected.

2.2. Statistical Analysis

Group comparisons between low BMD cases and healthy controls were carried out using independent t-tests implemented in Python 3.11.13. To identify genetic variants potentially linked to low BMD, logistic regression models were constructed using both SNP & Variation Suite (Golden Helix, Bozeman, MT, USA) and Persistent Linked INtegrated Kit version 1.9 (PLINK) [20]. Visualization of genome-wide association signals was performed through Manhattan and quantile-quantile (Q-Q) plots using Python 3.11.13. The chromosomal positions of significant SNPs were mapped using the PhenoGram visualization tool (http://visualization.ritchielab.org [accessed on 02 October 2018]) [21]. To reduce model-specific bias and enhance the confidence of common features, three models—linear discriminant analysis (LDA), random forest, and XGBoost model—were used to capture both linear (additive) effects and non-linear SNP interactions, thereby providing complementary information for variant prioritization. Machine learning analyses were applied as an auxiliary and exploratory approach to prioritize candidate SNP features associated with case status (OP = 1) versus controls (OP = 0). Linear discriminant analysis (LDA) was used to derive a one-dimensional discriminant component and to rank SNPs based on the absolute values of discriminant coefficients. Random forest and XGBoost classifiers were trained using an 80/20 train–test split, and SNPs were ranked by model-based feature importance scores. The top-ranked SNPs (top 10) from each model were visualized using PCA (tree-based models) and LDA projection (LDA), to illustrate separation patterns. To assess the potential functional impact of non-synonymous variants, we obtained functional prediction scores using the dbNSFP database [22], including the Sorting Intolerant from Tolerant (SIFT), Polymorphism Phenotyping v2, HumDiv-trained model/HumVar-trained model (PolyPhen-2 HDIV/HVAR), Protein Variation Effect Analyzer (PROVEAN), Rare Exome Variant Ensemble Learner (REVEL), and Combined Annotation-Dependent Depletion (CADD). Thresholds for predicting deleteriousness were set as follows: SIFT score < 0.05 [23], PolyPhen-2 HDIV/HVAR score > 0.85 for “probably damaging” [24], PROVEAN score < –2.5 [25], REVEL score > 0.5 [26], and CADD phred ≥ 20 [27]. To explore the evolutionary conservation of the overlapping SNPs, conservation scores were obtained using Genomic Evolutionary Rate Profiling (GERP++) [28], Phylogenetic p-value (phyloP) [29], and Phylogenetic Hidden Markov Model Conservation Score (phastCons) [30]. To explore protein–protein interaction networks, we used the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) v12.0 with a minimum required interaction score of 0.400, and clustering was performed using Markov Cluster Algorithm (MCL) clustering (inflation parameter = 2.0) (https://string-db.org [accessed on 16 August 2025]) [31]. Functional enrichment and pathway analysis of the identified genes were performed using Database for Annotation, Visualization, and Integrated Discovery (DAVID; https://davidbioinformatics.nih.gov [accessed on 16 August 2025]) [32]. Post hoc power analysis was conducted for selected SNPs highlighted in the main association results to evaluate whether the available sample size was sufficient to detect the observed effect sizes under a predefined genome-wide significance threshold (α = 1 × 10−5), using Python (version 3.11.13). Selection of SNPs was based on observed effect sizes (OR > 1.3 or < 0.7) and minor allele frequency (MAF) > 0.05. Minimum detectable odds ratios at 80% power were calculated using a Wald test–based normal approximation under an additive logistic regression model, incorporating sample size, case–control ratio, and minor allele frequency.

2.3. Linkage Disequilibrium Block Analysis and SNP Characterization

Genotype data obtained from the Illumina HumanExome BeadChip and the Affymetrix Axiom Exome Array were combined into a single working dataset. Before merging, both datasets underwent standard quality control using PLINK (version 1.9), including filters for call rate, minor allele frequency, and Hardy–Weinberg equilibrium. Because the two platforms use slightly different probe designs and allele encodings, we harmonized allele formats (including strand alignment) and updated SNP identifiers and genomic coordinates to ensure that variants aligned correctly across platforms. After preprocessing, pairwise linkage disequilibrium (LD) values (r2) were calculated using PLINK. LD blocks were then defined using a graph-based approach in which SNPs were treated as nodes and pairwise LD relationships (r2 ≥ 0.8) were represented as edges. LD blocks were identified as connected components within this SNP network, and SNP-to-block assignments were recorded for downstream block-level analyses. To focus on regions with potential biological relevance, we retained only LD blocks that included at least one SNP with an association p-value below 1 × 10−4. Within each selected block, the SNP with the smallest p-value was designated as the lead variant. For visualization and quality assessment, LD heatmaps were generated in R (version 4.3.3) by constructing block-wise r2 matrices, ordering SNPs by physical position, and visualizing within-block LD structure. This approach allowed inspection of LD consistency and block integrity across the two genotyping platforms. For each block, we also generated a LocusZoom-style plot to visualize the local association pattern around the lead SNP.

3. Results

3.1. Characteristics of Study Subjects

Age was significantly higher in the low BMD group, although age at menopause showed no significant difference. Alcohol and calcium intake were also comparable between groups. Notably, both body weight and BMI were significantly higher in the low BMD group. Moreover, while distal radius speed of sound (DR-SOS), distal radius T-score (DR-T), and distal radius Z-score (DR-Z) did not differ between groups, midshaft tibia speed of sound (MT-SOS), midshaft tibia T-score (MT-T), and midshaft tibia Z-score (MT-Z) demonstrated significant differences (Table 1).

3.2. Q-Q Plots and Manhattan Plots of Logistic Regression

Figure 2 shows the Q-Q plots and Manhattan plots from genome-wide association analyses using the Illumina Infinium HumanExome BeadChip and the Affymetrix Axiom Exome Array. Neither platform identified any SNPs meeting the genome-wide significance threshold (p < 5 × 10−8). In the Illumina Infinium HumanExome BeadChip data, one SNP exceeded the suggestive significance threshold (p < 1 × 10−5), and several SNPs displayed slight deviations from the expected line in the upper tail below the suggestive threshold, although the overall distribution closely matched the expected line. In the Affymetrix Axiom Exome Array data, association signals were generally lower, with few SNPs approaching or reaching the suggestive threshold, and distinct clusters were limited. The genomic inflation factors indicated minimal inflation for both platforms (λGC = 1.007 and 1.001, respectively).

3.3. Overlapping SNPs Across Two Genotyping Platforms (111 SNPs Across 70 Genes)

Figure 3a presents a chromosome map showing the locations of genes in which SNPs were commonly detected across both platforms, with each dot representing a distinct gene. Genes with the lowest p-values are highlighted in red. Table 2 lists the number of overlapping SNPs and their corresponding genes for each chromosome. Genes without overlapping SNPs were excluded. Notably, higher numbers of overlapping SNPs and genes were observed on chromosomes 1, 3, 6, 9, 11, and 12.
Figure 3b shows a scatter plot comparing the p-values (−log10 transformed) of SNPs analyzed from the same samples on the Illumina Infinium HumanExome BeadChip and the Affymetrix Axiom Exome Array. The red and blue dotted lines represent significance thresholds at p = 0.001 and p = 1 × 10−5, respectively. No SNPs reached the genome-wide significance threshold (p < 5 × 10−8) or the suggestive threshold (p < 1 × 10−5) on either platform.

3.4. Top SNP Selection via Multiple Machine Learning Models

In this study, we applied three machine learning techniques—LDA, random forest, and XGBoost—to assess the impact of SNPs with shared significance across two genome analysis platforms (Illumina Infinium HumanExome BeadChip and Affymetrix Axiom Exome Array) on osteoporosis risk. Table 3 presents the top 10 SNPs identified by each model for both platforms. SNPs with the highest coefficients or importance scores are listed for each model, enabling the identification of genes and SNPs common to multiple platforms and models. LDA results showed a relatively similar top SNP composition between platforms, whereas random forest and XGBoost yielded more divergent top SNP lists. Blank entries in the table indicate SNPs not directly linked to protein-coding genes.

3.5. Predicted Deleterious Non-Synonymous SNPs Identified Across Both Platforms by Multiple in Silico Algorithms

We analyzed overlapping SNPs identified across both platforms to identify SNPs with potential functional impact on proteins. Table 4 shows variants predicted to have deleterious or damaging effects by at least three of the following five algorithms (SIFT, PolyPhen2 HDIV/HVAR, PROVEAN, REVEL, and CADD). The prioritized SNPs were located in genes associated with including metabolic, cytoskeletal, and signal-transduction related genes, including ARMS2, CCDC92, NQO1, ZNF510, PTPRB, and DYNC2H1. Notably, CCDC92, NQO1, ZNF510, PTPRB and DYNC2H1 showed CADD phred scores ≥ 20, suggesting that these variants are among the top 1–5% of deleterious substitutions in the human genome. Several mutations, such as ARMS2 p.Ala69Ser and DYNC2H1 p.Arg2871Pro, were consistently predicted to be damaging by four or more algorithms.

3.6. Conservation Analysis

We performed a conservation analysis of overlapping SNPs identified across both platforms using GERP++, phyloP, and phastCons scores to evaluate their evolutionary conservation. As shown in Table 5, several SNPs, including DYNC2H1 (rs589623), NQO1 (rs1800566), and ZNF510 (rs2289651), exhibited high GERP++ scores (>2.0), indicating strong evolutionary constraint. PhyloP scores (both phyloP1 and phyloP4) were also markedly elevated for NQO1 (9.295 and 8.644, respectively) and DYNC2H1 (4.414 and 0.676), suggesting that these variants occur at highly conserved genomic positions across multiple species. Similarly, phastCons scores for DYNC2H1 and NQO1 approached 1.0, indicating strong probability of being in a conserved element.

3.7. Post Hoc Power and Minimum Detectable Effect Size Analysis

Post hoc power analysis was conducted to assess whether the sample size used in the SNP association analysis was sufficient to detect the observed effect sizes. Under a suggestive significance threshold (α = 1 × 10−5), most overlapping SNPs showed post hoc power values exceeding the predefined threshold. However, rs2584021 (PTPRB) exhibited low post hoc power (<0.20), suggesting limited statistical power for these variants. Among the non-redundant SNPs, rs2076212 (PNPLA3) was the only variant showing suggestive significance; however, its post hoc power was low (0.16), indicating that the observed association should be interpreted with caution due to the limited sample size. Nevertheless, although several SNPs showed statistically significant associations, none exceeded the minimum detectable odds ratio required for 80% power, indicating that the study remains underpowered to reliably detect small-to-moderate effect sizes given the modest sample size.

3.8. Protein–Protein Interactions and Functional Enrichment Analysis

We performed protein–protein interaction (PPI) analysis using the overlapping genes identified on both platforms. As shown in Figure 4a, the STRING network consisted of 65 nodes and 16 edges, with an average node degree of 0.492 and an average local clustering coefficient of 0.221. The PPI enrichment p-value was 0.122, indicating that the network did not contain significantly more interactions than would be expected by chance. Although not statistically enriched, several nodes showed multiple connections, which may imply limited functional grouping (Table 6). After MCL clustering, Figure 4b revealed functionally grouped modules. The largest cluster was enriched for genes involved in cytoskeletal organization and cellular projection–related processes. To further explore functional enrichment, subsets of genes were analyzed using STRING Gene Ontology categories. Figure 4c shows a cluster enriched for plasma membrane-bounded cell projection cytoplasm (GO:0032838; FDR = 9.41 × 10−11), including several microtubule-associated genes. Figure 4d shows another cluster related to distal axon (GO:0150034; FDR = 2.71 × 10−10).

3.9. LD Block Analysis and Cross-Platform Locus Characterization

Across the genome, we found ten LD blocks that included at least one SNP with p < 1 × 10−4. These blocks were spread across chromosomes 1, 2, 3, 5, 10, 11, and 12. Their structures were not uniform: some blocks consisted of only two SNPs, whereas others formed broader clusters containing more than 20 correlated variants. The list of SNPs included in each LD block is provided in Table A1. Chromosome 12 was particularly notable, as it contained four independent blocks corresponding to the NAV2, BICD1, CCDC92, and ZNF664 loci. To characterize these blocks in more detail, we visualized pairwise LD patterns using cross-platform LD heatmaps (Figure 5A). These heatmaps highlight how SNPs from the Illumina and Affymetrix exome arrays clustered together within each region, revealing both simple and complex LD structures depending on the locus. In parallel, regional association plots (Figure 5B) were generated to show the distribution of −log10(p) values surrounding each lead SNP, allowing us to examine the strength and breadth of association signals within each block.

4. Discussion

This study examined the genetic influence on the development of osteoporosis risk and low BMD in postmenopausal women. Table 1 presents the characteristics of the study subjects. Although the age at menopause was similar between groups, the prevalence of osteoporosis may increase with advancing age after menopause as with a previous study [33]. Underweight is generally associated with lower BMD and a higher risk of fracture; therefore, underweight individuals were excluded from this study [34]. Interestingly, the low BMD group showed higher body weight and BMI, which is contradictory to the general trend. However, both groups were classified as obese according to BMI criteria [35]. This finding may be related to the observation that DR-SOS, DR-T, and DR-Z did not differ significantly between groups, whereas MZ-SOS, MZ-T, and MZ-Z did. The radius is less influenced by walking or daily loading, whereas the tibia supports body weight, making bone loss in the tibia more pronounced. This functional difference may partly explain the variations in body weight and BMI [36]. Overall, postmenopausal women tend to show marked bone loss in the cortical midshaft tibia, whereas differences in the cancellous distal radius are relatively minimal or statistically insignificant.
Logistic regression analysis revealed that no SNPs met the genome-wide significance threshold [37]. However, one SNP exceeded the suggestive significance threshold (p < 1 × 10−5) on the Illumina Infinium HumanExome BeadChip. This SNP, rs2076212 of the PNPLA3 gene, is located within the coding region and represents a missense mutation (G115C). It was the only SNP to show statistical significance with an FDR of 0.016 (p < 0.05). PNPLA3 is a gene related to metabolism [38], and a previous study reported its association with osteoporosis [39]. That study also utilized KoGES data but included men, whereas the present study targeted only postmenopausal women. Notably, PNPLA3 is absent from the Affymetrix Axiom Exome Array, meaning it would not have been detected using Affymetrix alone. This highlights the advantage of expanded analysis using both platforms.
A comparatively larger number of overlapping SNPs and genes were observed on chromosomes 1, 3, 6, 9, 11, and 12. Many of these genes are involved in immune response (e.g., HLA-DOA) [40], cell signaling (e.g., DISC1, FGF12) [41,42], and bone metabolism (e.g., COL6A5) [43].
LDA revealed that certain genes, such as ATAD5, DNHD1, TNFSF15, and CCDC92, were recurrent across both platforms, suggesting that these variants may contribute to the discrimination between disease and control groups. In contrast, random forest and XGBoost model showed relatively low rates of common SNPs across platforms, and genes not identified by LDA appeared at the top of their rankings. This discrepancy likely reflects the fact that LDA selects variables contributing to classification based on linear discriminant criteria, whereas random forest and XGBoost consider nonlinear interactions. These findings indicate that the application of diverse models can be beneficial, since linear and non-linear methods may identify independent subsets of variants. Variants consistently detected across models are thus likely to represent particularly robust candidates [44]. SNPs that recur across platforms are more likely to be associated with disease and, therefore, should be prioritized in further analysis. Applying multiple models in parallel to identify overlapping variant candidates can provide more robust and reliable results [45].
To further narrow down the variants with potential biological relevance, we incorporated multiple in silico prediction tools. This approach allowed us to focus on SNPs that consistently showed signs of functional impact rather than relying on any single algorithm. Among the 110 candidates, six variants stood out by receiving deleterious predictions from several independent models, suggesting that they may meaningfully influence bone-related pathways. In particular, CCDC92 p.Ser70Cys and DYNC2H1 p.Arg2871Pro demonstrated strong and concordant signals, including high CADD phred scores (>20), which strengthens the likelihood that these variants play a functional role. Taken together, these findings present how integrating diverse computational assessments can help highlight variants that deserve closer biological or mechanistic investigation.
Conservation analysis provided additional support for prioritizing variants with potential functional significance. Variants with high GERP++, phyloP, and phastCons scores—particularly those in DYNC2H1 and NQO1—showed strong evolutionary constraint, reinforcing the pathogenic signals suggested by other in silico predictions. Among the prioritized SNPs, ZNF510 (rs2289651), ARMS2 (rs10490924), DYNC2H1 (rs589623), CCDC92 (rs11057401), PTPRB (rs2584021), and NQO1 (rs1800566) were consistently predicted to have deleterious effects. Several of these variants also ranked highly in our machine-learning models: CCDC92 in LDA, DYNC2H1 in XGBoost, and PTPRB in random forest. Together, these convergent lines of evidence suggest that these variants may contribute meaningfully to osteoporosis susceptibility in our cohort. In contrast, variants with weak statistical associations or low conservation scores may act through mechanisms less tied to evolutionary constraint or may reflect population-specific genetic variation.
Although the overall PPI enrichment was not statistically significant (PPI enrichment p > 0.05), a small cytoskeleton-related cluster could still be observed. Cluster analysis revealed an over-representation of genes related to cytoskeletal organization and cell projections, which may reflect the importance of mechanosensing and cell architecture regulation in bone remodeling. In particular, significant enrichment in the cytoplasmic region of membrane-bound cell projections and distal axon terminals suggests that osteogenic regulation may be mediated through pathways related to microtubule dynamics and cellular projection processes [46,47,48].
In this study, we investigated the genetic determinants of osteoporosis risk and low BMD in postmenopausal women using two complementary genotyping platforms and a multilayered analytic approach that included statistical testing, machine learning, in silico functional prediction, and protein–protein interaction analysis. Although genome-wide significance was not achieved, the suggestive association of rs2076212 (PNPLA3) highlights the benefit of integrating two platforms, as this variant was present only on the Illumina array and would not have been captured by a single-platform analysis.
Although PNPLA3 was the only variant to exceed the suggestive significance threshold (p < 1 × 10−5) in the single-SNP analyses across both platforms, integrating the datasets at the LD-block level revealed several additional loci with similarly strong association patterns. Because the Illumina and Affymetrix arrays differ substantially in variant coverage, many variants present on one platform are entirely absent from the other, making single-SNP testing inherently incomplete. By grouping correlated variants into LD-defined genomic regions using SNPs from both arrays, we were able to recover shared genetic signals that were not detectable from individual p-values alone. This LD-based approach noticeably improved the sensitivity of our analysis and helped reveal additional loci with potential biological relevance. In particular, four distinct regions on chromosome 12—NAV2, BICD1, CCDC92, and ZNF664—became apparent only when LD patterns were examined collectively across the two platforms, even though no individual SNP within these regions met strict significance thresholds on its own. These observations illustrate how LD-block integration can bridge platform-specific gaps, reduce discrepancies between arrays, and uncover extra genetic signals that may play a role in osteoporosis susceptibility.
A number of overlapping SNPs were repeatedly detected across chromosomes enriched in genes related to immune regulation, cell signaling, and bone metabolism. Among the machine-learning models, LDA showed greater cross-platform concordance than random forest and XGBoost model, suggesting that SNPs selected by multiple models and multiple platforms represent high-confidence candidates. Six variants (ARMS2, CCDC92, NQO1, ZNF510, PTPRB, and DYNC2H1) were consistently predicted to be functionally deleterious by ≥3 in silico tools, of which CCDC92 p.Ser70Cys and DYNC2H1 p.Arg2871Pro also showed high CADD scores and strong contributions in the machine-learning models. Furthermore, conservation scores supported the potential pathogenicity of DYNC2H1 and NQO1, reinforcing the likelihood of their involvement in osteoporosis risk. Interestingly, CCDC92 is known to be associated with adipose tissue biology and metabolic traits, whereas DYNC2H1 encodes a motor protein involved in intracellular transport and cytoskeletal regulation, suggesting that these genes may influence bone remodeling indirectly through mechanosensing and metabolic pathways [49,50,51]. In addition, NQO1 p.Pro187Ser, a variant involved in the oxidative stress response, may affect bone homeostasis through redox-mediated mechanisms [52]. PTPRB encodes a receptor-type tyrosine phosphatase that regulates angiogenesis and vascular integrity by modulating VEGFR2 signaling in endothelial cells [53]. These functional interpretations are consistent with our STRING clustering results, which revealed enrichment of cytoskeleton- and membrane-projection–related processes, indicating that mechanosensing and cell architecture could be key mediators of bone loss in postmenopausal women. In addition to the model-based results, the LD-block analysis offered an independent layer of evidence for several of these regions. Genes such as NAV2, BICD1, CCDC92, and ZNF664 appeared as clearly defined LD loci, even though none of the individual SNPs within these regions met genome-wide significance on their own. Notably, several genes highlighted by the machine-learning and in silico analyses—such as CCDC92, NQO1, and DYNC2H1—were also situated within these LD-structured regions. This overlap indicates that different analytic approaches were pointing to the same genomic areas, which strengthens confidence in their biological relevance. By grouping correlated variants into shared LD blocks, this approach also captured regional signals that would have been missed by single-SNP testing alone, allowing variants with modest p-values to be interpreted within a broader genetic context [54,55]. Together, these findings show how LD-block integration complements model-based prioritization and helps reveal additional loci that may contribute to osteoporosis risk.
Although this study was exploratory in nature, the identified candidate genes may have several potential clinical implications. First, when integrated with clinical and demographic factors, these genes may contribute to genetic risk assessment and stratification for low bone mineral density in postmenopausal women. Second, the associated genes are involved in biological pathways related to bone metabolism, mechanosensing, metabolic regulation, and cytoskeletal organization, making them promising candidates for targeted functional studies. Finally, the findings of this study may serve as a basis for hypothesis generation in future research aimed at evaluating biomarkers or therapeutic targets. However, additional replication studies and experimental validation are required before these findings can be translated into clinical practice or drug development.
This study combined two genotyping platforms and multiple analytic approaches—including machine learning, in silico prediction, conservation scoring, and network analysis—to strengthen the identification of biologically meaningful variants. Nevertheless, several limitations should be acknowledged. The final analytic sample size after quality control was modest, which may have limited statistical power to detect variants with modest effect sizes. Post hoc power analyses showed that many overlapping SNPs exceeded the predefined post hoc power threshold; however, several variants exhibited low post hoc power, and some statistically significant signals were observed despite limited power, indicating that these associations should be interpreted with caution. Overall, the modest sample size limited the study’s ability to reliably detect small-to-moderate genetic effects. This was further supported by minimum detectable effect size analyses, which showed that none of the observed associations exceeded the odds ratio required to achieve 80% power, emphasizing the need for validation in larger cohorts. Our study was conducted exclusively in Korean postmenopausal women. While this design inevitably limits the direct generalizability of the findings to other ancestral groups and does not fully exclude population-specific effects, the relative homogeneity of the cohort also represents a practical strength from a genetic epidemiology perspective. By reducing background genetic and environmental variability, this homogeneity can facilitate clearer detection of genetic signals within a defined population.
To address limitations related to the modest sample size and platform-specific SNP coverage, we integrated results from the two genotyping platforms using an LD-block–based framework. By analyzing correlated variants as regional units rather than relying solely on single-SNP tests, we were able to capture broader association patterns that were not apparent at the individual-variant level, particularly in genomic regions where the two arrays provided complementary, non-overlapping coverage. Notably, several regions identified through the LD-block analysis were concordant with loci highlighted by machine-learning and functional prediction approaches, providing convergent support across independent analytic strategies. Nevertheless, these findings should be regarded as hypothesis-generating. Validation in larger, independent cohorts—ideally incorporating multi-ethnic populations—and the use of higher-resolution genomic data will be essential to confirm the implicated loci and to more comprehensively define the genetic architecture of osteoporosis.

5. Conclusions

Taken together, this study highlights the genetic contributors to postmenopausal osteoporosis risk by integrating two genotyping platforms with LD-block analysis, machine learning, and in silico predictions. Although only PNPLA3 reached suggestive significance in single-SNP testing, LD-based refinement revealed additional loci—including NAV2, BICD1, CCDC92, and ZNF664—that were detectable only when cross-platform LD structure was considered. Convergent evidence across analytic layers prioritized variants such as CCDC92, DYNC2H1, and NQO1 as biologically meaningful candidates. These findings underscore the value of combining multi-platform data with structural and functional analyses and provide a focused set of variants for future validation.

Author Contributions

Conceptualization, S.K.K. and S.W.K.; Methodology, S.K.K.; Software, S.K.K.; Validation, S.K.K., S.S. and S.J.K.; Formal analysis, S.K.K.; Investigation, S.K.K., S.-J.H., S.I.S., J.K.L., G.K. and B.-J.C.; Resources, S.-J.H., S.I.S., J.K.L., G.K., B.-J.C. and J.Y.B.; Data curation, S.S. and S.J.K.; Writing—original draft preparation, S.K.K.; Writing—review and editing, S.W.K.; Visualization, S.K.K.; Supervision, S.W.K.; Project administration, S.W.K.; Funding acquisition, S.W.K. All authors have read and agreed to the published version of the manuscript.

Funding

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. NRF-2017R1C1B2012084). This study was conducted with bioresources from National Biobank of Korea, the Centers for Disease Control and Preven-tion, Republic of Korea (KBN-2018-058).

Institutional Review Board Statement

The study was conducted in accordance with the Declaration of Helsinki, and approved by the Institutional Review Board of the Dankook University (IRB no. 2018-08-004, Date: 10 September 2018).

Informed Consent Statement

As indicated in the IRB documents, this study was classified as a retrospective cohort/genetic data–based analysis, and therefore written informed consent was waived. The IRB approved this exemption and included a “Written Consent Waiver Statement” in the official documentation. Accordingly, no participant consent form was required, and none was collected.

Data Availability Statement

The raw genotype data used in this study were obtained from the Korean Genome and Epidemiology Study (KOGES) under a data-use license and cannot be shared publicly due to institutional and ethical restrictions. Summary-level statistical results generated during this study are available from the corresponding author upon reasonable request.

Acknowledgments

We thank the Korean Genome and Epidemiology Study (KoGES) and the Korean National Institute of Health for providing the cohort data used in this study. We also acknowledge the assistance of artificial intelligence tools during this work. ChatGPT 5 (OpenAI) and Gemini 2.5 flash (Google) supported coding processes, while ChatGPT 5 and QuillBot (https://quillbot.com) were used to assist in grammar checking and refining the English writing. The final responsibility for the content and interpretation of the results remains entirely with the authors.

Conflicts of Interest

The authors declare no conflicts of interest.

Abbreviations

The following abbreviations are used in this manuscript:
KoGESKorean Genome and Epidemiology Study
IRBInstitutional Review Board
SNPSingle Nucleotide Polymorphism
GWASGenome-Wide Association Study
LDLinkage Disequilibrium
MAFMinor Allele Frequency
HWEHardy–Weinberg Equilibrium
QCQuality Control
LDALinear Discriminant Analysis
PLINKPersistent Linked INtegrated Kit 
GERP++Genomic Evolutionary Rate Profiling
phyloPPhylogenetic p-value (phylogenetic conservation score)
phastConsPhylogenetic Hidden Markov Model Conservation Score
CADDCombined Annotation-Dependent Depletion
SIFTSorting Intolerant From Tolerant
PROVEANProtein Variation Effect Analyzer
REVELRare Exome Variant Ensemble Learner
STRINGSearch Tool for the Retrieval of Interacting Genes/Proteins
DAVIDDatabase for Annotation, Visualization and Integrated Discovery
MCLMarkov Cluster Algorithm
DR-SOSDistal Radius Speed of Sound
DR-TDistal Radius T-score
DR-ZDistal Radius Z-score
MT-SOSMidshaft Tibia Speed of Sound
MT-TMidshaft Tibia T-score
MT-ZMidshaft Tibia Z-score
BMIBody Mass Index
WHOWorld Health Organization 
BMDBone Mineral Density
Q-Q plotsQuantile-Quantile plots 

Appendix A

Table A1. Chromosomal LD Blocks and Their Constituent SNPs Identified Through Cross-Platform Integration.
Table A1. Chromosomal LD Blocks and Their Constituent SNPs Identified Through Cross-Platform Integration.
ChrChip IDrs IDCHRPosGeneFunction
1exm2249926rs558771871151230929PSMD4Silent
1SNP_A-2293176rs112047911151240542 
1SNP_A-1966584rs65875621151246241 
2SNP_A-1884547rs23569672193068151 
2SNP_A-2237714rs25922732193093850 
2SNP_A-4256617rs23569712193101110 
3exm359040rs37327653151090424MED12L, P2RY12Missense_R1210Q, Silent
3exm2265629rs98595383151090963MED12L, P2RY12Silent
3SNP_A-1974833rs38216633151100677MED12L 
3SNP_A-4205327rs109358403151101083MED12L 
3SNP_A-4217243rs172045013151114889MED12L 
3SNP_A-2166335rs172045083151115204MED12L 
3SNP_A-4203518rs46804063151116816MED12L 
3SNP_A-2041875rs22767653151131222MED12L 
5SNP_A-2221307rs893547592776972 
5SNP_A-1862456rs2344386592848652 
10SNP_A-2150516rs214847610122175555 
10SNP_A-2043377rs242071710122175979 
10SNP_A-4303428rs132666310122179526 
10SNP_A-2159301rs1088669010122213646PPAPDC1A
11SNP_A-2268822rs19144751128749414 
11SNP_A-1856716rs108353981128759826 
12SNP_A-2258849rs17982551232287259BICD1 
12exm2267339rs26084051232296621BICD1Silent
12SNP_A-2138255rs49316151232303400BICD1 
12SNP_A-1898535rs49316161232303456BICD1 
12SNP_A-4301892rs26305781232305787BICD1 
12SNP_A-4278412rs1619621232360803BICD1 
12SNP_A-4283755rs1619611232361233BICD1 
12SNP_A-1863901rs40177591277771495NAV2 
12SNP_A-2225499rs18808811277772555NAV2 
12SNP_A-2174026rs15270631277782433NAV2 
12SNP_A-1799650rs47613761277786244NAV2 
12SNP_A-1829924rs14650701277790549NAV2 
12SNP_A-2229564rs20111941277799416NAV2 
12SNP_A-1793840rs1105739412124407676DNAH10 
12SNP_A-1888303rs1105740112124427306CCDC92 
12exm1049349rs1105740112124427306CCDC92Missense_S70C
12SNP_A-2035335rs476521912124440110CCDC92 
12SNP_A-2209355rs730586412124441880CCDC92 
12SNP_A-1821027rs648891412124447841CCDC92 
12SNP_A-2163277rs476512712124460167ZNF664 
12exm-rs4765127rs476512712124460167ZNF664Silent
12SNP_A-2288649rs1231111412124460703ZNF664 
12SNP_A-2267281rs476552812124462254ZNF664 
12SNP_A-2296376rs1105740812124464836ZNF664 
12SNP_A-4238292rs797861012124468572ZNF664 
12SNP_A-1787908rs731196912124470333ZNF664 
12SNP_A-2079815rs731123312124475940ZNF664 
12SNP_A-2194556rs476514812124478637ZNF664 
12SNP_A-1867568rs476556812124479161ZNF664 
12SNP_A-4204952rs1105740912124479331ZNF664 
12SNP_A-1961399rs797548212124481690ZNF664 
12SNP_A-2058919rs118741512124491529ZNF664 
12SNP_A-2222659rs730705312124494540ZNF664 

References

  1. Kanis, J.A. Assessment of fracture risk and its application to screening for postmenopausal osteoporosis: Synopsis of a WHO report. WHO Study Group. Osteoporos. Int. 1994, 4, 368–381. [Google Scholar] [CrossRef]
  2. Cummings, S.R.; Melton, L.J. Epidemiology and outcomes of osteoporotic fractures. Lancet 2002, 359, 1761–1767. [Google Scholar] [CrossRef]
  3. Rachner, T.D.; Khosla, S.; Hofbauer, L.C. Osteoporosis: Now and the future. Lancet 2011, 377, 1276–1287. [Google Scholar] [CrossRef]
  4. Harris, M.; Nguyen, T.V.; Howard, G.M.; Kelly, P.J.; Eisman, J.A. Genetic and environmental correlations between bone formation and bone mineral density: A twin study. Bone 1998, 22, 141–145. [Google Scholar] [CrossRef] [PubMed]
  5. Trajanoska, K.; Morris, J.A.; Oei, L.; Zheng, H.F.; Evans, D.M.; Kiel, D.P.; Ohlsson, C.; Richards, J.B.; Rivadeneira, F.; GEFOS/GENOMOS consortium and the 23andMe research team; et al. Assessment of the genetic and clinical determinants of fracture risk: Genome wide association and mendelian randomisation study. BMJ 2018, 362, k3225. [Google Scholar] [CrossRef] [PubMed]
  6. Choi, H.J.; Park, H.; Zhang, L.; Kim, J.H.; Kim, Y.A.; Yang, J.Y.; Pei, Y.F.; Tian, Q.; Shen, H.; Hwang, J.Y.; et al. Genome-wide association study in East Asians suggests UHMK1 as a novel bone mineral density susceptibility gene. Bone 2016, 91, 113–121. [Google Scholar] [CrossRef]
  7. Bae, Y.S.; Im, S.W.; Kang, M.S.; Kim, J.H.; Lee, S.H.; Cho, B.L.; Park, J.H.; Nam, Y.S.; Son, H.Y.; Yang, S.D.; et al. Genome-Wide Association Study of Bone Mineral Density in Korean Men. Genom. Inf. 2016, 14, 62–68. [Google Scholar] [CrossRef] [PubMed]
  8. Maher, B. Personal genomes: The case of the missing heritability. Nature 2008, 456, 18–21. [Google Scholar] [CrossRef] [PubMed]
  9. Lee, S.; Abecasis, G.R.; Boehnke, M.; Lin, X. Rare-variant association analysis: Study designs and statistical tests. Am. J. Hum. Genet. 2014, 95, 5–23. [Google Scholar] [CrossRef]
  10. Page, C.M.; Baranzini, S.E.; Mevik, B.H.; Bos, S.D.; Harbo, H.F.; Andreassen, B.K. Assessing the Power of Exome Chips. PLoS ONE 2015, 10, e0139642. [Google Scholar] [CrossRef]
  11. Mizrahi-Man, O.; Woehrmann, M.H.; Webster, T.A.; Gollub, J.; Bivol, A.; Keeble, S.M.; Aull, K.H.; Mittal, A.; Roter, A.H.; Wong, B.A.; et al. Novel genotyping algorithms for rare variants significantly improve the accuracy of Applied Biosystems Axiom array genotyping calls: Retrospective evaluation of UK Biobank array data. PLoS ONE 2022, 17, e0277680. [Google Scholar] [CrossRef]
  12. Sun, T.H.; Shao, Y.J.; Mao, C.L.; Hung, M.N.; Lo, Y.Y.; Ko, T.M.; Hsiao, T.H. A Novel Quality-Control Procedure to Improve the Accuracy of Rare Variant Calling in SNP Arrays. Front. Genet. 2021, 12, 736390. [Google Scholar] [CrossRef]
  13. Weedon, M.N.; Jackson, L.; Harrison, J.W.; Ruth, K.S.; Tyrrell, J.; Hattersley, A.T.; Wright, C.F. Use of SNP chips to detect rare pathogenic variants: Retrospective, population based diagnostic evaluation. BMJ 2021, 372, n214, Correction in BMJ 2021, 372, n792. https://doi.org/10.1136/bmj.n792. [Google Scholar] [CrossRef]
  14. Visscher, P.M.; Brown, M.A.; McCarthy, M.I.; Yang, J. Five years of GWAS discovery. Am. J. Hum. Genet. 2012, 90, 7–24. [Google Scholar] [CrossRef]
  15. Gabriel, S.B.; Schaffner, S.F.; Nguyen, H.; Moore, J.M.; Roy, J.; Blumenstiel, B.; Higgins, J.; DeFelice, M.; Lochner, A.; Faggart, M.; et al. The structure of haplotype blocks in the human genome. Science 2002, 296, 2225–2229. [Google Scholar] [CrossRef]
  16. Marin, F.; Gonzalez-Macias, J.; Diez-Perez, A.; Palma, S.; Delgado-Rodriguez, M. Relationship between bone quantitative ultrasound and fractures: A meta-analysis. J. Bone Min. Res. 2006, 21, 1126–1135. [Google Scholar] [CrossRef]
  17. Park, S.J.; Jung, J.H.; Kim, M.S.; Lee, H.J. High dairy products intake reduces osteoporosis risk in Korean postmenopausal women: A 4 year follow-up study. Nutr. Res. Pr. 2018, 12, 436–442. [Google Scholar] [CrossRef] [PubMed]
  18. Morin, S.N.; Leslie, W.D.; Schousboe, J.T. Osteoporosis: A Review. JAMA 2025, 334, 894–907. [Google Scholar] [CrossRef]
  19. Anderson, C.A.; Pettersson, F.H.; Clarke, G.M.; Cardon, L.R.; Morris, A.P.; Zondervan, K.T. Data quality control in genetic case-control association studies. Nat. Protoc. 2010, 5, 1564–1573. [Google Scholar] [CrossRef] [PubMed]
  20. Purcell, S.; Neale, B.; Todd-Brown, K.; Thomas, L.; Ferreira, M.A.; Bender, D.; Maller, J.; Sklar, P.; de Bakker, P.I.; Daly, M.J.; et al. PLINK: A tool set for whole-genome association and population-based linkage analyses. Am. J. Hum. Genet. 2007, 81, 559–575. [Google Scholar] [CrossRef] [PubMed]
  21. Wolfe, D.; Dudek, S.; Ritchie, M.D.; Pendergrass, S.A. Visualizing genomic information across chromosomes with PhenoGram. BioData Min. 2013, 6, 18. [Google Scholar] [CrossRef]
  22. Li, C.; Zhi, D.; Wang, K.; Liu, X. MetaRNN: Differentiating rare pathogenic and rare benign missense SNVs and InDels using deep learning. Genome Med. 2022, 14, 115. [Google Scholar] [CrossRef]
  23. Sim, N.L.; Kumar, P.; Hu, J.; Henikoff, S.; Schneider, G.; Ng, P.C. SIFT web server: Predicting effects of amino acid substitutions on proteins. Nucleic Acids Res. 2012, 40, W452–W457. [Google Scholar] [CrossRef]
  24. Adzhubei, I.A.; Schmidt, S.; Peshkin, L.; Ramensky, V.E.; Gerasimova, A.; Bork, P.; Kondrashov, A.S.; Sunyaev, S.R. A method and server for predicting damaging missense mutations. Nat. Methods 2010, 7, 248–249. [Google Scholar] [CrossRef]
  25. Choi, Y.; Chan, A.P. PROVEAN web server: A tool to predict the functional effect of amino acid substitutions and indels. Bioinformatics 2015, 31, 2745–2747. [Google Scholar] [CrossRef] [PubMed]
  26. Ioannidis, N.M.; Rothstein, J.H.; Pejaver, V.; Middha, S.; McDonnell, S.K.; Baheti, S.; Musolf, A.; Li, Q.; Holzinger, E.; Karyadi, D.; et al. REVEL: An Ensemble Method for Predicting the Pathogenicity of Rare Missense Variants. Am. J. Hum. Genet. 2016, 99, 877–885. [Google Scholar] [CrossRef] [PubMed]
  27. Rentzsch, P.; Witten, D.; Cooper, G.M.; Shendure, J.; Kircher, M. CADD: Predicting the deleteriousness of variants throughout the human genome. Nucleic Acids Res. 2019, 47, D886–D894. [Google Scholar] [CrossRef] [PubMed]
  28. Davydov, E.V.; Goode, D.L.; Sirota, M.; Cooper, G.M.; Sidow, A.; Batzoglou, S. Identifying a high fraction of the human genome to be under selective constraint using GERP++. PLoS Comput. Biol. 2010, 6, e1001025. [Google Scholar] [CrossRef]
  29. Pollard, K.S.; Hubisz, M.J.; Rosenbloom, K.R.; Siepel, A. Detection of nonneutral substitution rates on mammalian phylogenies. Genome Res. 2010, 20, 110–121. [Google Scholar] [CrossRef]
  30. Garcia, F.A.O.; de Andrade, E.S.; Palmero, E.I. Insights on variant analysis in silico tools for pathogenicity prediction. Front. Genet. 2022, 13, 1010327. [Google Scholar] [CrossRef]
  31. Szklarczyk, D.; Kirsch, R.; Koutrouli, M.; Nastou, K.; Mehryary, F.; Hachilif, R.; Gable, A.L.; Fang, T.; Doncheva, N.T.; Pyysalo, S.; et al. The STRING database in 2023: Protein-protein association networks and functional enrichment analyses for any sequenced genome of interest. Nucleic Acids Res. 2023, 51, D638–D646. [Google Scholar] [CrossRef] [PubMed]
  32. Sherman, B.T.; Hao, M.; Qiu, J.; Jiao, X.; Baseler, M.W.; Lane, H.C.; Imamichi, T.; Chang, W. DAVID: A web server for functional enrichment analysis and functional annotation of gene lists (2021 update). Nucleic Acids Res. 2022, 50, W216–W221. [Google Scholar] [CrossRef] [PubMed]
  33. Kanis, J.A.; Gluer, C.C. An update on the diagnosis and assessment of osteoporosis with densitometry. Osteoporos. Int. 2000, 11, 192–202. [Google Scholar] [CrossRef]
  34. Ravn, P.; Cizza, G.; Bjarnason, N.H.; Thompson, D.; Daley, M.; Wasnich, R.D.; McClung, M.; Hosking, D.; Yates, A.J.; Christiansen, C. Low body mass index is an important risk factor for low bone mass and increased bone loss in early postmenopausal women. Early Postmenopausal Intervention Cohort (EPIC) study group. J. Bone Min. Res. 1999, 14, 1622–1627. [Google Scholar] [CrossRef]
  35. Consultation, W.H.O.E. Appropriate body-mass index for Asian populations and its implications for policy and intervention strategies. Lancet 2004, 363, 157–163. [Google Scholar] [CrossRef]
  36. Seeman, E.; Delmas, P.D. Bone quality--the material and structural basis of bone strength and fragility. N. Engl. J. Med. 2006, 354, 2250–2261. [Google Scholar] [CrossRef]
  37. Pe’er, I.; Yelensky, R.; Altshuler, D.; Daly, M.J. Estimation of the multiple testing burden for genomewide association studies of nearly all common variants. Genet. Epidemiol. 2008, 32, 381–385. [Google Scholar] [CrossRef]
  38. Li, T.H.; Huang, Y.S.; Ma, C.C.; Tsai, S.Y.; Tsai, H.C.; Yeh, H.Y.; Shen, H.C.; Hong, S.Y.; Su, C.W.; Yang, H.I.; et al. GCKR Polymorphisms Increase the Risks of Low Bone Mineral Density in Young and Non-Obese Patients With MASLD and Hyperuricemia. Kaohsiung J. Med. Sci. 2025, 41, e70017. [Google Scholar] [CrossRef] [PubMed]
  39. Hong, E.P.; Rhee, K.H.; Kim, D.H.; Park, J.W. Identification of pleiotropic genetic variants affecting osteoporosis risk in a Korean elderly cohort. J. Bone Min. Metab. 2019, 37, 43–52. [Google Scholar] [CrossRef]
  40. Okada, Y.; Suzuki, A.; Ikari, K.; Terao, C.; Kochi, Y.; Ohmura, K.; Higasa, K.; Akiyama, M.; Ashikawa, K.; Kanai, M.; et al. Contribution of a Non-classical HLA Gene, HLA-DOA, to the Risk of Rheumatoid Arthritis. Am. J. Hum. Genet. 2016, 99, 366–374. [Google Scholar] [CrossRef]
  41. Brandon, N.J.; Sawa, A. Linking neurodevelopmental and synaptic theories of mental illness through DISC1. Nat. Rev. Neurosci. 2011, 12, 707–722. [Google Scholar] [CrossRef] [PubMed]
  42. Goldfarb, M.; Schoorlemmer, J.; Williams, A.; Diwakar, S.; Wang, Q.; Huang, X.; Giza, J.; Tchetchik, D.; Kelley, K.; Vega, A.; et al. Fibroblast growth factor homologous factors control neuronal excitability through modulation of voltage-gated sodium channels. Neuron 2007, 55, 449–463. [Google Scholar] [CrossRef]
  43. Wang, X.; Pandey, A.K.; Mulligan, M.K.; Williams, E.G.; Mozhui, K.; Li, Z.; Jovaisaite, V.; Quarles, L.D.; Xiao, Z.; Huang, J.; et al. Joint mouse-human phenome-wide association to test gene function and disease risk. Nat. Commun. 2016, 7, 10464. [Google Scholar] [CrossRef]
  44. Abraham, G.; Inouye, M. Genomic risk prediction of complex human disease and its clinical application. Curr. Opin. Genet. Dev. 2015, 33, 10–16. [Google Scholar] [CrossRef]
  45. Libbrecht, M.W.; Noble, W.S. Machine learning applications in genetics and genomics. Nat. Rev. Genet. 2015, 16, 321–332. [Google Scholar] [CrossRef]
  46. Ma, M.; Chen, X.; Lu, L.; Yuan, F.; Zeng, W.; Luo, S.; Yin, F.; Cai, J. Identification of crucial genes related to postmenopausal osteoporosis using gene expression profiling. Aging Clin. Exp. Res. 2016, 28, 1067–1074. [Google Scholar] [CrossRef] [PubMed]
  47. Wang, N.; Tytell, J.D.; Ingber, D.E. Mechanotransduction at a distance: Mechanically coupling the extracellular matrix with the nucleus. Nat. Rev. Mol. Cell Biol. 2009, 10, 75–82. [Google Scholar] [CrossRef] [PubMed]
  48. Bonewald, L.F. Mechanosensation and Transduction in Osteocytes. Bonekey Osteovision 2006, 3, 7–15. [Google Scholar] [CrossRef]
  49. Ren, L.; Du, W.; Song, D.; Lu, H.; Hamblin, M.H.; Wang, C.; Du, C.; Fan, G.C.; Becker, R.C.; Fan, Y. Genetic ablation of diabetes-associated gene Ccdc92 reduces obesity and insulin resistance in mice. iScience 2023, 26, 105769. [Google Scholar] [CrossRef]
  50. Roberts, A.J. Emerging mechanisms of dynein transport in the cytoplasm versus the cilium. Biochem. Soc. Trans. 2018, 46, 967–982. [Google Scholar] [CrossRef]
  51. Yuan, X.; Yang, S. Cilia/Ift protein and motor -related bone diseases and mouse models. Front. Biosci. (Landmark Ed.) 2015, 20, 515–555. [Google Scholar] [CrossRef]
  52. Martin, N.J.; Collier, A.C.; Bowen, L.D.; Pritsos, K.L.; Goodrich, G.G.; Arger, K.; Cutter, G.; Pritsos, C.A. Polymorphisms in the NQO1, GSTT and GSTM genes are associated with coronary heart disease and biomarkers of oxidative stress. Mutat. Res. 2009, 674, 93–100. [Google Scholar] [CrossRef]
  53. Hayashi, M.; Majumdar, A.; Li, X.; Adler, J.; Sun, Z.; Vertuani, S.; Hellberg, C.; Mellberg, S.; Koch, S.; Dimberg, A.; et al. VE-PTP regulates VEGFR2 activity in stalk cells to establish endothelial cell polarity and lumen formation. Nat. Commun. 2013, 4, 1672. [Google Scholar] [CrossRef] [PubMed]
  54. Genomes Project, C.; Auton, A.; Brooks, L.D.; Durbin, R.M.; Garrison, E.P.; Kang, H.M.; Korbel, J.O.; Marchini, J.L.; McCarthy, S.; McVean, G.A.; et al. A global reference for human genetic variation. Nature 2015, 526, 68–74. [Google Scholar] [CrossRef] [PubMed]
  55. Visscher, P.M.; Wray, N.R.; Zhang, Q.; Sklar, P.; McCarthy, M.I.; Brown, M.A.; Yang, J. 10 Years of GWAS Discovery: Biology, Function, and Translation. Am. J. Hum. Genet. 2017, 101, 5–22. [Google Scholar] [CrossRef] [PubMed]
Diagnostics 16 00153 g001
Figure 1. Flowchart of participant selection and SNP filtering process. Flowchart summarizing the selection of postmenopausal women from the Ansan and Anseong cohorts of the KoGES, 3rd period, and the subsequent SNP filtering steps. After applying exclusion criteria, participants were genotyped using two independent platforms (Illumina Infinium HumanExome BeadChip and Affymetrix Axiom Exome Array). SNPs were filtered by statistical significance, allele frequency, and HWE, and overlapping variants were identified for further analysis.
Figure 1. Flowchart of participant selection and SNP filtering process. Flowchart summarizing the selection of postmenopausal women from the Ansan and Anseong cohorts of the KoGES, 3rd period, and the subsequent SNP filtering steps. After applying exclusion criteria, participants were genotyped using two independent platforms (Illumina Infinium HumanExome BeadChip and Affymetrix Axiom Exome Array). SNPs were filtered by statistical significance, allele frequency, and HWE, and overlapping variants were identified for further analysis.
Diagnostics 16 00153 g001
Diagnostics 16 00153 g002
Figure 2. Q-Q plots and Manhattan plots of logistic regression p-values for low BMD (cases vs. controls). (a) Illumina Infinium HumanExome BeadChip Q–Q plot, (b) Illumina Infinium HumanExome BeadChip Manhattan plot, (c) Affymetrix Axiom Exome Array Q-Q plot, and (d) Affymetrix Axiom Exome Array Manhattan plot. Dots represent the –log10(p) values of SNPs, alternately colored blue and red for each chromosome. Horizontal lines indicate p = 1 × 10−3 (green dotted line), p = 1 × 10−5 (orange dotted line), and p = 5 × 10−8 (red solid line; genome-wide cutoff). No SNP exceeds the genome-wide cutoff in any panel. In the Illumina exome plot (b), some SNPs exceed the suggestive threshold (p < 1 × 10−5), whereas in the Affymetrix plot (d), the signals are generally weaker. Q-Q plots are largely consistent with the expected line (y = x) except for deviations in the tails, suggesting that systematic bias is minimal. Blue and red dots in (b,d) indicate alternating chromosomes for visual distinction only.
Figure 2. Q-Q plots and Manhattan plots of logistic regression p-values for low BMD (cases vs. controls). (a) Illumina Infinium HumanExome BeadChip Q–Q plot, (b) Illumina Infinium HumanExome BeadChip Manhattan plot, (c) Affymetrix Axiom Exome Array Q-Q plot, and (d) Affymetrix Axiom Exome Array Manhattan plot. Dots represent the –log10(p) values of SNPs, alternately colored blue and red for each chromosome. Horizontal lines indicate p = 1 × 10−3 (green dotted line), p = 1 × 10−5 (orange dotted line), and p = 5 × 10−8 (red solid line; genome-wide cutoff). No SNP exceeds the genome-wide cutoff in any panel. In the Illumina exome plot (b), some SNPs exceed the suggestive threshold (p < 1 × 10−5), whereas in the Affymetrix plot (d), the signals are generally weaker. Q-Q plots are largely consistent with the expected line (y = x) except for deviations in the tails, suggesting that systematic bias is minimal. Blue and red dots in (b,d) indicate alternating chromosomes for visual distinction only.
Diagnostics 16 00153 g002
Diagnostics 16 00153 g003
Figure 3. Overlapping variants identified across two genotyping platforms. (a) Chromosomal ideogram showing the genomic locations of genes harboring overlapping SNPs detected in both the Illumina Infinium HumanExome BeadChip and the Affymetrix Axiom Exome Array. Each colored circle represents a unique gene, and red-labeled genes indicate those with the lowest p-values. (b) Scatter plot comparing –log10(p) values for overlapping SNPs between the two platforms. Red dashed lines indicate a nominal significance level (p = 0.001), and the blue dashed line marks the suggestive significance threshold (p = 1 × 10−5). No SNP reached the genome-wide significance threshold (p < 5 × 10−8); however, rs2076212 exhibited suggestive significance on the Illumina platform. The different dot colors in Figure 3a are used solely to visually distinguish SNP loci.
Figure 3. Overlapping variants identified across two genotyping platforms. (a) Chromosomal ideogram showing the genomic locations of genes harboring overlapping SNPs detected in both the Illumina Infinium HumanExome BeadChip and the Affymetrix Axiom Exome Array. Each colored circle represents a unique gene, and red-labeled genes indicate those with the lowest p-values. (b) Scatter plot comparing –log10(p) values for overlapping SNPs between the two platforms. Red dashed lines indicate a nominal significance level (p = 0.001), and the blue dashed line marks the suggestive significance threshold (p = 1 × 10−5). No SNP reached the genome-wide significance threshold (p < 5 × 10−8); however, rs2076212 exhibited suggestive significance on the Illumina platform. The different dot colors in Figure 3a are used solely to visually distinguish SNP loci.
Diagnostics 16 00153 g003
Diagnostics 16 00153 g004
Figure 4. Protein–protein interaction (PPI) network and functional clustering of overlapping genes identified on both platforms. (a) STRING PPI network, (b) MCL-based clustered network showing functional grouping of genes (inflation = 2.0), (c) Subnetwork enriched for plasma membrane-bounded cell projection cytoplasm (FDR = 9.41 × 10−11), (d) Subnetwork enriched for distal axon (GO:0150034; FDR = 2.71 × 10−10). Nodes represent proteins encoded by overlapping genes, and colors indicate cluster membership.
Figure 4. Protein–protein interaction (PPI) network and functional clustering of overlapping genes identified on both platforms. (a) STRING PPI network, (b) MCL-based clustered network showing functional grouping of genes (inflation = 2.0), (c) Subnetwork enriched for plasma membrane-bounded cell projection cytoplasm (FDR = 9.41 × 10−11), (d) Subnetwork enriched for distal axon (GO:0150034; FDR = 2.71 × 10−10). Nodes represent proteins encoded by overlapping genes, and colors indicate cluster membership.
Diagnostics 16 00153 g004
Diagnostics 16 00153 g005
Figure 5. Cross-platform LD block architecture and regional association patterns for SNPs. (A) LD heatmaps for each block containing a SNP with p < 1 × 10−4. SNPs are ordered by genomic position, colored by platform (Affymetrix Axiom Exome Array: blue; Illumina Infinium HumanExome BeadChip: red), with lead SNPs outlined. These heatmaps summarize local LD patterns and cross-platform concordance. (B) Regional association plots showing −log10(p) values within ±150 kb of each lead SNP. Points are colored by LD (r2) to illustrate LD decay, and the vertical dashed line marks the lead SNP. These plots highlight the peak signal and its surrounding LD structure. In the LD heatmaps, the circled marker indicates the lead SNP. In the regional association plots, each point represents a SNP and the vertical dashed line indicates the genomic position of the lead SNP.
Figure 5. Cross-platform LD block architecture and regional association patterns for SNPs. (A) LD heatmaps for each block containing a SNP with p < 1 × 10−4. SNPs are ordered by genomic position, colored by platform (Affymetrix Axiom Exome Array: blue; Illumina Infinium HumanExome BeadChip: red), with lead SNPs outlined. These heatmaps summarize local LD patterns and cross-platform concordance. (B) Regional association plots showing −log10(p) values within ±150 kb of each lead SNP. Points are colored by LD (r2) to illustrate LD decay, and the vertical dashed line marks the lead SNP. These plots highlight the peak signal and its surrounding LD structure. In the LD heatmaps, the circled marker indicates the lead SNP. In the regional association plots, each point represents a SNP and the vertical dashed line indicates the genomic position of the lead SNP.
Diagnostics 16 00153 g005
Table 1. Basic characteristics of the postmenopausal women in this study. DR-T/DR-Z and MT-T/MT-Z denote QUS-derived T- and Z-scores from SOS measurements at the distal radius (DR) and midshaft tibia (MT), respectively. These indices were used for the operational definition of osteoporosis risk/low BMD and are not directly comparable to DXA-based WHO/ISCD diagnostic criteria.
Table 1. Basic characteristics of the postmenopausal women in this study. DR-T/DR-Z and MT-T/MT-Z denote QUS-derived T- and Z-scores from SOS measurements at the distal radius (DR) and midshaft tibia (MT), respectively. These indices were used for the operational definition of osteoporosis risk/low BMD and are not directly comparable to DXA-based WHO/ISCD diagnostic criteria.
ControlLow BMDp-Value
age (years)53.99 ± 4.5558.09 ± 4.240.000
age at menopausal (years)49.59 ± 2.8749.47 ± 3.8>0.05
weight (kg)64.05 ± 5.1565.54 ± 6.880.034
BMI (kg/m2)25.98 ± 2.3527.57 ± 2.880.000
alcohol consumption (g/day)0.59 ± 1.40.57 ± 1.55>0.05
calcium consumption (mg/day)<1000<1000-
DR-SOS (m/s)4200.76 ± 118.274000.31 ± 177.36>0.05
DR-T (m/s)0.3 ± 0.9−1.29 ± 1.46>0.05
DR-Z (m/s)1.12 ± 1.090.1 ± 1.39>0.05
MT-SOS (m/s)3936.63 ± 61.313625.35 ± 122.150.000
MT-T (m/s)−0.2 ± 0.59−3.24 ± 1.190.000
MT-Z (m/s)0.59 ± 0.73−1.92 ± 1.320.000
Table 2. Overlapping SNPs and Genes by Chromosome.
Table 2. Overlapping SNPs and Genes by Chromosome.
ChrnGene
18KIF1B, ZYG11A, PROK1, DISC1
26NRXN1, MARCO, CWC22, MAP2
311C3orf77, COL6A5, ATR, CPB1, GFM1, FGF12
43DGKQ, SORBS2
56PDZD2, C7, IQGAP2
616NRM, NOTCH4, HLA-DOA, BTBD9, THEMIS, LAMA2, SYNE1
72ASNS
85SLC18A1, C8orf86, UBXN2B, FAM135B
910KDM4C, ACER2, ZNF510, TNFSF15, CERCAM, SETX
104KIAA1217, NRG3, SH3PXD2A
116OSBPL5, DNHD1, MMP13, DYNC2H1
1212CLECL1, SLC4A8, MYO1A, PTPRB, C12orf64, CCDC63, WDR66, GPR81, CCDC92, ZNF664
134SLC46A3
143OTX2
152TNFAIP8L3, LINS
163TRAP1, NQO1, ADAMTS18
173DLG4, ATAD5
180 
191 
203RNF114
211URB1
X0 
Y0 
Chr: chromosome; n: number.
Table 3. Top 10 SNPs identified by LDA, random forest, and XGBoost models for Illumina Infinium HumanExome BeadChip and Affymetrix Axiom Exome Array.
Table 3. Top 10 SNPs identified by LDA, random forest, and XGBoost models for Illumina Infinium HumanExome BeadChip and Affymetrix Axiom Exome Array.
LDARandom ForestXGBoost
rsIDGeneCoefficientrsIDGeneImportancersIDGeneImportance
Illumina Infinium HumanExome BeadChiprs11657270ATAD52.61222rs2584021PTPRB0.012048rs11248060DGKQ0.01788
rs4263839TNFSF150.424226rs9554742 0.011968rs8134971URB10.017365
rs4758423DNHD10.38457rs11124754 0.011813rs1049674ASNS0.016935
rs11057401CCDC920.378416rs10109439FAM135B0.011802rs6478108TNFSF150.016367
rs3129304HLA-DOA0.260162rs557135 0.011747rs10253361 0.015085
rs1049674ASNS−0.254866rs4406360 0.011317rs4679621 0.014259
rs4633449DNHD1−0.356667rs4947122 0.011187rs6556756 0.014067
rs6478108TNFSF15−0.433349rs6556756 0.011181rs589623DYNC2H10.013548
rs4765127ZNF664−0.436117rs1169081WDR660.011081rs10964136ACER20.013402
rs3816780ATAD5−2.322621rs10490924 0.011063rs763318 0.013336
Affymetrix Axiom Exome Arrayrs11057401CCDC921.099642rs2008344TRAP10.013578rs10124818 0.021613
rs11657270ATAD50.777047rs7305599SLC4A80.013159rs2229032ATR0.019727
rs4633449DNHD10.437324rs7514102PROK10.01289rs629648THEMIS0.018916
rs4263839TNFSF150.393153rs4758540OSBPL50.012731rs4633449DNHD10.015453
rs11247226LINS0.315041rs10253361 0.012213rs353372 0.014661
rs6478108TNFSF15−0.459254rs8134971URB10.012191rs6033098 0.014627
rs2229032ATR−0.525691rs10109439FAM135B0.011849rs6795735 0.014511
rs4758423DNHD1−0.567479rs12033321 0.011474rs763318 0.013953
rs3816780ATAD5−0.896848rs1009850CERCAM0.011446rs10253361 0.013829
rs4765127ZNF664−1.089067rs1169081WDR660.011368rs10748869NRG30.013601
Table 4. Overlapping SNPs predicted to have deleterious functional effects by multiple in silico algorithms.
Table 4. Overlapping SNPs predicted to have deleterious functional effects by multiple in silico algorithms.
SNP IDChrPosGeneAmino Acid ChangeSIFTPolyphen2 HDIVPolyphen2 HVARPROVEANREVELCADD
ScorePredScorePredScorePredScorePredScorePhred
rs1049092410122,454,932ARMS2p.Ala69Ser0D0.994D0.957D−2.63D0.06115.87
rs1105740112123,942,759CCDC92p.Ser70Cys0.005D1D0.971D−2.44N0.16423.3
rs18005661669,711,242NQO1p.Pro187Ser0.032D0.438B0.167B−7.39D0.36624
rs2289651996,774,789ZNF510p.Gln43Pro0.01D0.838P0.202B−3.65D0.12822.2
rs25840211270,635,953PTPRBp.Asp57Tyr0.004D0.978D0.77P−1.12N0.21420.7
rs58962311103,211,861DYNC2H1p.Arg2871Pro0.015D0.991D0.964D−4.2D0.30127.4
Chr: chromosome; Pos: position; Pred: prediction; Phred: phred-scaled C-score.
Table 5. Evolutionary Conservation Scores (GERP++, phyloP, and phastCons) for Overlapping SNPs Identified Across Both Platforms.
Table 5. Evolutionary Conservation Scores (GERP++, phyloP, and phastCons) for Overlapping SNPs Identified Across Both Platforms.
SNP IDChrPosGeneGERP++phyloP
(V)		phyloP
(M)		phyloP
(P)		phastCons
(V)		phastCons
(M)		phastCons
(P)
rs9284879344,243,092TOPAZ12.961.4072.166−0.1060.92810.975
rs2289651996,774,789ZNF5101.521.944−2.1740.6650.99900.963
rs1049092410122,454,932ARMS20.9980.215 0.6180.0060.0080.025
rs58962311103,211,861DYNC2H15.764.414 0.676110.997
rs1105740112123,942,759CCDC923.443.0051.7630.661110.995
rs25840211270,635,953PTPRB3.920.980.8480.5990.7630.4460.947
rs18005661669,711,242NQO15.419.2958.6440.676110.997
Chr: chromosome; Pos: position; (V): vertebrate; (M): mammalian; (P): primate.
Table 6. Functional clusters of overlapping genes identified by STRING.
Table 6. Functional clusters of overlapping genes identified by STRING.
ClusterGene CountPrimary DescriptionProtein Names
19Kinesin bindingKIF1B, DLG4, PTPRB, MAP2, SYNE1, DISC1, NRG3, NRXN1, PDZD2
24miscellaneousSETX, THEMIS, ATR, ATAD5
34miscellaneousTRAP1, GFM1, KIAA1217, CWC22
42Mixed, incl. Domain of unknown function DUF4537, and CCDC92/74, N-terminalCCDC92, ZNF664
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Kim, S.K.; Hong, S.-J.; Song, S.I.; Lee, J.K.; Kim, G.; Choi, B.-J.; Seon, S.; Kim, S.J.; Ban, J.Y.; Kang, S.W. Integrated Cross-Platform Analysis Reveals Candidate Variants and Linkage Disequilibrium-Defined Loci Associated with Osteoporosis in Korean Postmenopausal Women. Diagnostics 2026, 16, 153. https://doi.org/10.3390/diagnostics16010153

AMA Style

Kim SK, Hong S-J, Song SI, Lee JK, Kim G, Choi B-J, Seon S, Kim SJ, Ban JY, Kang SW. Integrated Cross-Platform Analysis Reveals Candidate Variants and Linkage Disequilibrium-Defined Loci Associated with Osteoporosis in Korean Postmenopausal Women. Diagnostics. 2026; 16(1):153. https://doi.org/10.3390/diagnostics16010153

Chicago/Turabian Style

Kim, Su Kang, Seoung-Jin Hong, Seung Il Song, Jeong Keun Lee, Gyutae Kim, Byung-Joon Choi, Suyun Seon, Seung Jun Kim, Ju Yeon Ban, and Sang Wook Kang. 2026. "Integrated Cross-Platform Analysis Reveals Candidate Variants and Linkage Disequilibrium-Defined Loci Associated with Osteoporosis in Korean Postmenopausal Women" Diagnostics 16, no. 1: 153. https://doi.org/10.3390/diagnostics16010153

APA Style

Kim, S. K., Hong, S.-J., Song, S. I., Lee, J. K., Kim, G., Choi, B.-J., Seon, S., Kim, S. J., Ban, J. Y., & Kang, S. W. (2026). Integrated Cross-Platform Analysis Reveals Candidate Variants and Linkage Disequilibrium-Defined Loci Associated with Osteoporosis in Korean Postmenopausal Women. Diagnostics, 16(1), 153. https://doi.org/10.3390/diagnostics16010153

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop