ABSTRACTHarmonizing osteoporosis‐related data across multiple data sets is essential for improving the accuracy and generalizability of bone mineral density (BMD) assessments. This study developed a harmonization framework to standardize phenotypic and genomic variables across three major US osteoporosis data sets: GDBF, GWAS, and NHANES. We standardized key phenotypic variables (BMD, body mass index (BMI), age, sex, and race/ethnicity) using cohort‐specific data dictionaries and applied multiple imputations by chained equations (MICEs) to manage missing data. Genomic data were harmonized using principal component analysis (PCA)‐based batch effect corrections. Residual regression methods were applied to standardize BMD values. The effectiveness of harmonization on BMD prediction was evaluated using generalized estimating equations (GEEs) and mixed‐effects models. Post‐harmonization, inter‐study variability in BMI was significantly reduced (Ω2 = 0.0028), and BMD associations with covariates remained consistent across data sets. Harmonized models showed improved predictive performance, with explained variance in BMD increasing (R2 = 0.14%). PCA confirmed the effective alignment of genetic data, reducing batch effects and improving cross‐study compatibility. This study demonstrates the feasibility and effectiveness of harmonizing phenotypic and genomic data for osteoporosis research. The harmonization framework enhances BMD prediction accuracy, supports more inclusive osteoporosis risk assessment, and improves the integration of multi‐cohort data sets for future research. These findings highlight the potential of data harmonization in advancing precision medicine for osteoporosis prevention and management.

Herpes simplex viruses (HSV) have been detected within the synovial joint cavity, a secluded area of inflammation that may harbor etiological agents. However, the role of HSV-1/2 infection in arthritis pathogenesis remains ambiguous. In this study, we integrate cross-sectional epidemiology and genetic associations to elucidate their relationships and uncover causal mechanisms. We analyzed cross-sectional data from 18,292 NHANES participants (1999-2016) using multivariable-adjusted logistic regression to assess associations between anti-HSV-1/2 IgG seropositivity and arthritis-related risks. Complementary analyses included linkage disequilibrium score regression (LDSC) and bidirectional Mendelian randomization (MR) using genetic instruments for anti-HSV IgG levels to explore genetic correlations and infer causality. Initial observational findings demonstrated significant positive associations between HSV-1/2 IgG seropositivity and arthritis risk (all p < 0.001); however, these associations lost significance after multivariable adjustment. Notably, after multivariable adjustment, subtype analyses revealed that HSV-2 IgG seropositivity was linked to increased risks of rheumatoid arthritis (RA) (OR: 1.40, 95% CI: 1.04-1.88) and osteoarthritis (OA) (OR: 1.39, 95% CI: 1.07-1.81), while HSV-1 IgG seropositivity correlated with an unknown-arthritis subtype (OR: 1.38, 95% CI: 1.08-1.75). Moreover, MR analyses uncovered divergent causal effects: anti-HSV-1 IgG levels were protective against OA (OR: 0.90, 95% CI: 0.82-0.98), whereas anti-HSV-2 IgG levels modestly increased OA risk (OR: 1.05, 95% CI: 1.01-1.09). No reverse causation or genetic correlation was observed. This study's innovative integration of epidemiological and genetic methodologies not only clarifies the distinct roles of HSV sub-types in arthritis but also identifies HSV-2 as a potential causal factor in OA, thereby opening new avenues for therapeutic targeting.

ABSTRACTGenetic risk scores (GRS) are crucial tools for estimating an individual's genetic liability to various traits and diseases, computed as a weighted sum of trait‐associated allele counts. Traditionally, GRS models assume additive, linear effects of risk variants. However, complex traits often involve nonadditive interactions, such as epistasis, which are not captured by these conventional methods. In this study, we investigate the use of random forest (RF) models as a model‐free approach for constructing GRS, leveraging RF's capacity to capture complex, nonlinear interactions among genetic variants. Specifically, we introduce two new RF‐based GRS strategies to boost RF performance and to incorporate base data information if available, including (1) ctRF, which optimizes linkage disequilibrium (LD) clumping and p‐value thresholds within RF; and (2) wRF, which adjusts the chance of SNP inclusion in tree nodes based on their association strength. Through simulation studies and real data applications of Alzheimer's disease, body mass index, and atopy, we find that ctRF consistently outperforms other RF‐based methods and classical additive models when traits exhibit complex genetic architectures. Additionally, incorporating informative base data into RF‐GRS construction can enhance predictive accuracy. Our findings suggest that RF‐based GRS can effectively capture intricate genetic interactions, and offer a robust alternative to traditional GRS methods, especially for complex traits with nonlinear genetic effects.

The polygenic score (PGS) holds promise for motivating preventive behavioral changes. However, no clinically validated standardization methodology currently exists. Here, we demonstrate the efficacy of a "reference-based" approach for standardization. This method uses the PGS distribution in the general population as a reference for normalization and percentile determination; however, it has not been validated. We investigated three potential influences on PGS computation: (1) the size of the reference population, (2) biases associated with different genotyping platforms, and (3) inclusion of kinship ties within the reference group. Our results indicate that the reference size affects the bootstrap estimate of standard error for PGS percentiles, peaking around the 50th percentile and diminishing at extreme percentiles (1st or 100th). Discrepancies between genotyping platforms, such as different microarrays and whole-genome sequencing, resulted in deviations in PGS (p < 0.05 in Kolmogorov-Smirnov test). However, these deviations were reduced to a nonsignificant level using shared genetic variants in the calculations when the ancestry of the samples and reference were matched. This approach recovered approximately 9.6% of the positive predictive value of PGS by naïve genotype. Our results provide fundamental insights for establishing clinical guidelines for implementing PGS to communicate reliable risks to individuals.

ABSTRACTBiological datasets often consist of thousands or millions of variables, e.g. genetic variants or biomarkers, and when sample sizes are large it is common to find many associated with an outcome of interest, for example, disease risk in a GWAS, at high levels of statistical significance, but with very small effects. The False Discovery Rate (FDR) is used to identify effects of interest based on ranking variables according to their statistical significance. Here, we develop a complementary measure to the FDR, the priorityFDR, that ranks variables by a combination of effect size and significance, allowing further prioritisation among a set of variables that pass a significance or FDR threshold. Applying to the largest GWAS of type 1 diabetes to date (15,573 cases and 158,408 controls), we identified 26 independent genetic associations, including two newly‐reported loci, with qualitatively lower priorityFDRs than the remaining 175 signals. We detected putatively causal type 1 diabetes risk genes using Mendelian Randomisation, and found that these were located disproportionately close to low priorityFDR signals (p = 0.005), as were genes in the IL‐2 pathway (p = 0.003). Selecting variables on both effect size and significance can lead to improved prioritisation for mechanistic follow‐up studies from genetic and other large biological datasets.

Transcriptome-wide association studies (TWAS) have been widely used to identify thousands of likely causal genes for diseases and complex traits using predicted expression models. However, most existing TWAS methods rely on gene expression alone and overlook other regulatory mechanisms of gene expression, including DNA methylation and splicing, that contribute to the genetic basis of these complex traits and diseases. Here we introduce a multi-omics method that integrates gene expression, DNA methylation, and splicing data to improve the identification of associated genes with our traits of interest. Through simulations and by analyzing genome-wide association study (GWAS) summary statistics for 24 complex traits, we show that our integrated method, which leverages these complementary omics biomarkers, achieves higher statistical power, and improves the accuracy of likely causal gene identification in blood tissues over individual omics methods. Finally, we apply our integrated model to a lung cancer GWAS data set, demonstrating the integrated models improved identification of prioritized genes for lung cancer risk.

ABSTRACTTranscriptome‐wide association studies (TWAS) aim to uncover genotype–phenotype relationships through a two‐stage procedure: predicting gene expression from genotypes using an expression quantitative trait locus (eQTL) data set, then testing the predicted expression for trait associations. Accurate gene expression prediction in stage 1 is crucial, as it directly impacts the power to identify associations in stage 2. Currently, the first stage of such studies is primarily conducted using linear models like elastic net regression, which fail to capture the nonlinear relationships inherent in biological systems. Deep learning methods have the potential to model such nonlinear effects, but have yet to demonstrably outperform linear methods at this task. To address this gap, we propose a new deep learning architecture to predict gene expression from genotypic variation across individuals. Our method utilizes a learnable input scaling layer in conjunction with a convolutional encoder to capture nonlinear effects and higher‐order interactions without compromising on interpretability. We further augment this approach to allow for parameter sharing across multiple networks, enabling us to utilize prior information for individual variants in the form of functional annotations. Evaluations on real‐world genomic data show that our method consistently outperforms elastic net regression across a large set of heritable genes. Furthermore, our model statistically significantly improved predictive performance by leveraging functional annotations, whereas elastic net regression failed to show equivalent gains when using the same information, suggesting that our method can capture nonlinear functional information beyond the capability of linear models.

In most Proteome-Wide Association Studies (PWAS), variants near the protein-coding gene (±1 Mb), also known as cis single nucleotide polymorphisms(SNPs), are used to predict protein levels, which are then tested for association with phenotypes. However, proteins can be regulated through variants outside of the cis region. An intermediate GWAS step to identify protein quantitative trait loci (pQTL) allows for the inclusion of trans SNPs outside the cis region in protein-level prediction models. Here, we assess the prediction of 540 proteins in 1002 individuals from the Women's Health Initiative (WHI), split equally into a GWAS set, an elastic net training set, and a testing set. We compared the testing r2 between measured and predicted protein levels using this proposed approach, to the testing r2 using only cis SNPs. The two methods usually resulted in similar testing r2, but some proteins showed a significant increase in testing r2 with our method. For example, for cartilage acidic protein 1, the testing r2 increased from 0.101 to 0.351. We also demonstrate reproducible findings for predicted protein association with lipid and blood cell traits in WHI participants without proteomics data and in UK Biobank utilizing our PWAS weights.