Mingxuan CAI

Spatial multi-omics data offer a powerful framework for integrating diverse molecular profiles while maintaining the spatial organization of cells. However, inherent variations in data quality and noise levels across different modalities pose significant challenges to accurate integration and analyses. In this paper, we introduce CANDIES, which leverages a conditional diffusion model and contrastive learning to effectively denoise and integrate spatial multi-omics data. With our innovative model and algorithm designs, CANDIES not only enhances the quality of spatial multi-omics data, but also yields a unified and comprehensive joint representation, thereby empowering many downstream analyses. We conduct extensive evaluations on diverse synthetic and real datasets, including MISAR-seq data from the mouse brain, spatial CITE-seq data from human skin biopsy tissue, spatial Mux-seq, and spatial ATAC-RNA-seq data from the mouse embryo, and 10\\times Visium data from human lymph nodes. CANDIES shows superior performance on various downstream tasks, including denoising, spatial domain identification, spatiotemporal trajectory reconstruction, and spatial association mapping for complex human traits. In particular, we show that CANDIES representations can be integrated with the rich resources from genome-wide association studies (GWASs), allowing the spatial domains to be linked with complex human traits, yielding spatially resolved interpretations of complex traits in their relevant tissues.

Fine-mapping aims to prioritize causal variants underlying complex traits by accounting for the linkage disequilibrium of GWAS risk locus. The expanding resources of functional annotations serve as auxiliary evidence to improve the power of fine-mapping. However, existing fine-mapping methods tend to generate many false positive results when integrating a large number of annotations. In this study, we propose a unified method to integrate high-dimensional functional annotations with fine-mapping (Funmap). Funmap can effectively improve the power of fine-mapping by borrowing information from hundreds of functional annotations. Meanwhile, it relates the annotation to the causal probability with a random effects model that avoids the over-fitting issue, thereby producing a well-controlled false positive rate. Paired with a fast algorithm, Funmap enables scalable integration of a large number of annotations to facilitate prioritizing multiple causal SNPs. Our simulations demonstrate that Funmap is the only method that produces well-calibrated FDR under the setting of high-dimensional annotations while achieving better or comparable power gains as compared to existing methods. By integrating GWASs of 4 lipid traits with 187 functional annotations, Funmap consistently identified more variants that can be replicated in an independent cohort, achieving 15.5%-26.2% improvement over the runner-up in terms of replication rate.

In various practical situations, matrix factorization methods suffer from poor data quality, such as high data sparsity and low signal-to-noise ratio (SNR). Here we consider a matrix factorization problem by utilizing auxiliary information, which is massively available in real applications, to overcome the challenges caused by poor data quality. Unlike existing methods that mainly rely on simple linear models to combine auxiliary information with the main data matrix, we propose to integrate gradient boosted trees in the probabilistic matrix factorization framework to effec- tively leverage auxiliary information (MFAI). Thus, MFAI naturally inherits several salient features of gradient boosted trees, such as the capability of flexibly modeling nonlinear relationships, and robustness to irrelevant features and missing values in auxiliary information. The parameters in MAFI can be automatically determined under the empirical Bayes framework, making it adaptive to the utilization of aux- iliary information and immune to overfitting. Moreover, MFAI is computationally efficient and scalable to large-scale datasets by exploiting variational inference. We demonstrate the advantages of MFAI through comprehensive numerical results from simulation studies and real data analysis. Our approach is implemented in the R package mfair available at https://github.com/YangLabHKUST/mfair.

Fine-mapping prioritizes risk variants identified by genome-wide association studies (GWASs), serving as a critical step to uncover biological mechanisms underlying complex traits. However, several major challenges still remain for existing fine-mapping methods. First, the strong linkage disequilibrium among variants can limit the statistical power and resolution of fine-mapping. Second, it is computationally expensive to simultaneously search for multiple causal variants. Third, the confounding bias hidden in GWAS summary statistics can produce spurious signals. To address these challenges, we develop a statistical method for cross-population fine-mapping (XMAP) by leveraging genetic diversity and accounting for confounding bias. By using cross-population GWAS summary statistics from global biobanks and genomic consortia, we show that XMAP can achieve greater statistical power, better control of false positive rate, and substantially higher computational efficiency for identifying multiple causal signals, compared to existing methods. Importantly, we show that the output of XMAP can be integrated with single-cell datasets, which greatly improves the interpretation of putative causal variants in their cellular context at single-cell resolution.

Over the past two decades, genome-wide association studies (GWASs) have successfully advanced our understanding of genetic basis of complex traits. Despite the fruitful discovery of GWASs, most GWAS samples are collected from European populations, and these GWASs are often criticized for their lack of ancestry diversity. Trans-ancestry association mapping (TRAM) offers an exciting opportunity to fill the gap of disparities in genetic studies between non-Europeans and Europeans. Here we propose a statistical method, LOG-TRAM, to leverage the local genetic architecture for TRAM. By using biobank-scale datasets, we showed that LOG-TRAM can greatly improve the statistical power of identifying risk variants in under-represented populations while producing well-calibrated p-values. We applied LOG-TRAM to the GWAS summary statistics of 29 complex traits/diseases from Biobank Japan (BBJ) and UK Biobank (UKBB), and achieved substantial gains in power (the effective sample sizes increased by 49% in average compared to the BBJ GWASs) and effective correction of confounding biases compared to existing methods. Finally, we demonstrated that LOG-TRAM can be successfully applied to identify ancestry-specific loci and the LOG-TRAM output can be further used for construction of more accurate polygenic risk scores (PRSs) in under-represented populations.Competing Interest StatementThe authors have declared no competing interest.

As increasing sample sizes from genome-wide association studies (GWASs), polygenic risk scores (PRSs) have shown great potential in personalized medicine with disease risk prediction, prevention and treatment. However, the PRS constructed using European samples becomes less accurate when it is applied to individuals from non-European populations. It is an urgent task to improve the accuracy of PRSs in under-represented populations, such as African populations and East Asian populations.In this paper, we propose a cross-population and cross-phenotype (XPXP) method for construction of PRSs in under-represented populations. XPXP can construct accurate PRSs by leveraging biobank-scale datasets in European populations and multiple GWASs of genetically correlated phenotypes. XPXP also allows to incorporate population-specific and phenotype-specific effects, and thus further improves the accuracy of PRS. Through comprehensive simulation studies and real data analysis, we demonstrated that our XPXP outperformed existing PRS approaches. We showed that the height PRSs constructed by XPXP achieved 9\% and 18\% improvement over the runner-up method in terms of predicted R2 in East Asian and African populations, respectively. We also showed that XPXP substantially improved the stratification ability in identifying individuals at high genetic risk of Type 2 Diabetes.The XPXP software and all analysis code are available at github.com/YangLabHKUST/XPXPSupplementary data are available at Bioinformatics online.

We present a unified statistical framework (XPA) to improve the prediction accuracy of human traits using multi-ancestry genetic data. Paired with innovations in data structure and algorithm design, our framework is highly scalable, with both computational cost and memory storage linear to the sample size and number of predictors. In practice, XPA can analyze 3 million variants from 430K samples with only 385 Gb memory usage in 54.5 hours. In a Chinese cohort, our method achieves 7.3%-198.0% accuracy gain for height prediction in terms of R2 compared to existing methods.

Many genetic variants affect phenotypes by regulating the gene expression level. We develop a statistical model, IGREX, to quantify the impact of genetically regulated expression on various human traits and inform trait-relevant tissue types. Efficient parameter expanded EM (PX-EM) algorithm and Method of Moments are adopted to optimize computational efficiency.

In this article, we consider a Bayesian bi-level variable selection problem in high-dimensional regressions. In many practical situations, it is natural to assign group membership to each predictor. Examples include that genetic variants can be grouped at the gene level and a covariate from different tasks naturally forms a group. Thus, it is of interest to select important groups as well as important members from those groups. The existing Markov chain Monte Carlo methods are often computationally intensive and not scalable to large datasets. To address this problem, we consider variational inference for bi-level variable selection. In contrast to the commonly used mean-field approximation, we propose a hierarchical factorization to approximate the posterior distribution, by using the structure of bi-level variable selection. Moreover, we develop a computationally efficient and fully parallelizable algorithm based on this variational approximation. We further extend the developed method to model datasets from multitask learning. The comprehensive numerical results from both simulation studies and real data analysis demonstrate the advantages of BIVAS for variable selection, parameter estimation, and computational efficiency over existing methods. The method is implemented in R package “bivas” available at https:// github.com/ mxcai/ bivas. Supplementary materials for this article are available online.