Biology subjects

Conti, D. V.

Publications and source records attributed to Conti, D. V..

8 recordsLinked to original sources

bioRxiv2025

Clustering of Omic Data Using Semi-Supervised Transfer Learning for Gaussian Mixture Models via Natural-Gradient Variational Inference: Method and Applications to Bulk and Single-Cell Transcriptomics

MotivationHigh-dimensional omics data are typically measured on limited sample sizes, which challenges model-based clustering methods such as Gaussian mixture models, often leading to instability and poor generalization under complex mixture structures. To address these limitations, we developed Praxis-BGM, a natural-gradient variational inference framework for Gaussian mixture models that enables semi-supervised transfer learning by incorporating an informative prior Gaussian mixture model derived from large-scale reference data with robust cluster structures. This prior can encode cluster-specific means, covariance structures, and structural connectivity patterns, and is updated using the target data with variational inference to improve clustering in small-sample settings. ResultsWe derived natural-gradient updates for standard parameters and assess feature-level contributions to posterior clustering via Bayes Factors. Implemented in Python library JAX for accelerator-oriented computation, Praxis-BGM is computationally efficient and scalable. Across extensive simulations and two real-world applications--breast cancer bulk transcriptomics for subtype recovery and single-cell transcriptomics for cross-platform label transfer--Praxis-BGM improves posterior clustering performance, stability, and biological interpretability, even when priors are partially mismatched. Availability and ImplementationPraxis-BGM is freely available at https://github.com/ContiLab-usc/Praxis-BGM, and an archival version is available on Zenodo at https://doi.org/10.5281/zenodo.19657680. Contactqiranjia@usc.edu Supplementary InformationSupplementary materials are available with the manuscript submission.

bioinformatics↗

bioRxiv2025

An Efficient Lasso Framework for Admixture-Aware Polygenic Scores

Polygenic scores (PGS) have promising clinical applications for risk stratification, disease screening, and personalized medicine. However, most PGS are trained on predominantly European ancestry cohorts and have limited portability to external populations. While cross-population PGS methods have demonstrated greater generalizability than single-ancestry PGS, they fail to properly account for individuals with recent admixture between continental ancestry groups. GAUDI is a recently proposed PGS method which overcomes this gap by leveraging local ancestry to estimate ancestry-specific effects, penalizing but allowing ancestry-differential effects. However, the modified fused LASSO approach used by GAUDI is computationally expensive and does not readily accommodate more than two-way admixture. To address these limitations, we introduce HAUDI, an efficient LASSO framework for admixed PGS construction. HAUDI re-parameterizes the GAUDI model as a standard LASSO problem, allowing for extension to multi-way admixture settings and far superior computational speed than GAUDI. In extensive simulations, HAUDI compares favorably to GAUDI while dramatically reducing computation time. In real data applications, HAUDI uniformly out-performs GAUDI across 18 clinical phenotypes, including total triglycerides (TG), C-reactive protein (CRP), and mean corpuscular hemoglobin concentration (MCHC), and shows substantial benefits over an ancestry-agnostic PGS for white blood cell count (WBC) and chronic kidney disease (CKD).

genomics↗

bioRxiv2023

Estimating heritability explained by local ancestry and evaluating stratification bias in admixture mapping from summary statistics

The heritability explained by local ancestry markers in an admixed population [Formula] provides crucial insight into the genetic architecture of a complex disease or trait. Estimation of [Formula] can be susceptible to biases due to population structure in ancestral populations. Here, we present a novel approach, Heritability estimation from Admixture Mapping Summary STAtistics (HAMSTA), which uses summary statistics from admixture mapping to infer heritability explained by local ancestry while adjusting for biases due to ancestral stratification. Through extensive simulations, we demonstrate that HAMSTA [Formula] estimates are approximately unbiased and are robust to ancestral stratification compared to existing approaches. In the presence of ancestral stratification, we show a HAMSTA-derived sampling scheme provides a calibrated family-wise error rate (FWER) of [~]5% for admixture mapping, unlike existing FWER estimation approaches. We apply HAMSTA to 20 quantitative phenotypes of up to 15,988 self-reported African American individuals in the Population Architecture using Genomics and Epidemiology (PAGE) study. We observe [Formula] in the 20 phenotypes range from 0.0025 to 0.033 (mean [Formula]), which translates to [Formula] ranging from 0.062 to 0.85 (mean [Formula]). Across these phenotypes we find little evidence of inflation due to ancestral population stratification in current admixture mapping studies (mean inflation factor of 0.99 +/-0.001). Overall, HAMSTA provides a fast and powerful approach to estimate genome-wide heritability and evaluate biases in test statistics of admixture mapping studies.

genetics↗

bioRxiv2022

Fine-Mapping and Credible Set Construction using a Multipopulation Joint Analysis of Marginal Summary Statistics from Genome-wide Association Studies

Recent advancement in Genome-wide Association Studies (GWAS) comes from not only increasingly larger sample sizes but also the shifted focus towards underrepresented populations. Multi-population GWAS may increase power to detect novel risk variants and improve fine-mapping resolution by leveraging evidence from diverse populations and accounting for the difference in linkage disequilibrium (LD) across ethnic groups. Here, we expand upon our previous approach for single-population fine-mapping through Joint Analysis of Marginal SNP Effects (JAM) to a multi-population analysis (mJAM). Under the assumption that true causal variants are common across studies, we implement a novel version of JAM that conditions on multiple SNPs while explicitly incorporating the different LD structures across populations. The mJAM framework can be used to first select index variants using the mJAM likelihood with any feature selection approach. In addition, we present a novel approach leveraging the ideas of mediation to construct credible sets for these index variants. Construction of such credible sets can be performed given any existing index variants. We illustrate the implementation of the mJAM likelihood through two implementations: mJAM-SuSiE (a Bayesian approach) and mJAM-Forward selection. Through simulation studies based on realistic effect sizes and levels of LD, we demonstrated that mJAM performs better than other existing multi-ethnic methods for constructing concise credible sets that include the underlying causal variants. In real data examples taken from the most recent multi-population prostate cancer GWAS, we showed several practical advantages of mJAM over other existing methods.

genetics↗

bioRxiv2022

Inverted genomic regions between reference genome builds in humans impact imputation accuracy and decrease the power of association testing

Over the last two decades, the human reference genome has undergone multiple updates as we complete a linear representation of our genome. There are two versions of human references currently used in the biomedical literature, GRCh37/hg19 and GRCh38, and conversions between these versions are critical for quality control, imputation, and association analysis. In the present study we show that genomic coordinates for single nucleotide variants (SNVs) in regions inverted between different builds of the reference genome are erroneously converted by the TOPMed imputation server. Depending on the array type, we estimate the inappropriate conversions of variant coordinates would occur in about 2-5 Mb of the genome. Independent of the imputation server, however, in-house conversions of variant coordinates are routine yet impact of mapped variants to these inverted regions and its downstream analytical consequences are underrecognized. We observed that errors for palindromic variants in these inverted regions cannot be detected by standard quality control procedures and destabilize the local haplotype structure, leading to loss of imputation accuracy and power in association analyses. Though only a small proportion of the genome is affected, we show that these regions include important disease susceptibility variants that would be lost due to poor imputation. For example, we show that a known locus associated with prostate cancer on chr10 would have its association P-value drop from 2.86x10-7 to 0.0011 in a case-control analysis of 20,286 Africans and African Americans (10,643 cases and 9,643 controls). We devise a straight-forward heuristic based on the popular tool liftOver that can easily detect and correct these variants in the inverted regions between genome builds to locally improve imputation accuracy.

genomics↗

bioRxiv2022

Multi-ancestry fine-mapping improves precision to identify causal genes in transcriptome-wide association studies

Transcriptome-wide association studies (TWAS) are a powerful approach to identify genes whose expression associates with complex disease risk. However, non-causal genes can exhibit association signals due to confounding by linkage disequilibrium patterns (LD) and eQTL pleiotropy at genomic risk regions which necessitates fine-mapping of TWAS signals. Here, we present MA-FOCUS, a multi-ancestry framework for the improved identification of genes underlying traits of interest. We demonstrate that by leveraging differences in ancestry-specific patterns of LD and eQTL signals, MA-FOCUS consistently outperforms single-ancestry fine-mapping approaches with equivalent total sample size across multiple metrics. We perform 15 blood trait TWAS using genome-wide summary statistics (average NEA=511k, NAA=13k) and lymphoblastoid cell line eQTL data from cohorts of primarily European and African continental ancestries. We recapitulate evidence demonstrating shared genetic architectures for eQTL and blood traits between the two ancestry groups and observe that gene-level effects correlate 20% more strongly across ancestries compared with SNP-level effects. We perform fine-mapping using MA-FOCUS and find evidence that genes at TWAS risk regions are more likely to be shared across ancestries rather than ancestry-specific. Using multiple lines of evidence to validate our findings, we find gene sets produced by MA-FOCUS are more enriched in hematopoietic categories compared to alternative approaches (P = 1.73 x 10-16). Our work demonstrates that including, and appropriately accounting for, genetic diversity can drive deeper insights into the genetic architecture of complex traits.

genetics↗

bioRxiv2020

A Hierarchical Approach Using Marginal Summary Statistics for Multiple Intermediates in a Mendelian Randomization or Transcriptome Analysis

BackgroundPrevious research has demonstrated the usefulness of hierarchical modeling for incorporating a flexible array of prior information in genetic association studies. When this prior information consists of effect estimates from association analyses of single nucleotide polymorphisms (SNP)-intermediate or SNP-gene expression, a hierarchical model is equivalent to a two-stage instrumental or transcriptome-wide association study (TWAS) analysis, respectively. MethodsWe propose to extend our previous approach for the joint analysis of marginal summary statistics (JAM) to incorporate prior information via a hierarchical model (hJAM). In this framework, the use of appropriate effect estimates as prior information yields an analysis similar to Mendelian Randomization (MR) and TWAS approaches such as FUSION and S-PrediXcan. However, hJAM is applicable to multiple correlated SNPs and multiple correlated intermediates to yield conditional estimates of effect for the intermediate on the outcome, thus providing advantages over alternative approaches. ResultsWe investigate the performance of hJAM in comparison to existing MR approaches (inverse-variance weighted MR and multivariate MR) and existing TWAS approaches (S-PrediXcan) for effect estimation, type-I error and empirical power. We apply hJAM to two examples: estimating the conditional effects of body mass index and type 2 diabetes on myocardial infarction and estimating the effects of the expressions of gene NUCKS1 and PM20D1 on the risk of prostate cancer. ConclusionsAcross numerous causal simulation scenarios, we demonstrate that hJAM is unbiased, maintains correct type-I error and has increased power. Key MessagesO_LIMendelian randomization and transcriptome-wide association studies (TWAS) can be viewed as similar approaches via a hierarchical model. C_LIO_LIThe hierarchal joint analysis of marginal summary statistics (hJAM) is a multivariate Mendelian randomization approach which offers a simple way to address the pleiotropy bias that is introduced by genetic variants associated with multiple risk factors or expressions of genes. C_LIO_LIhJAM incorporates the linkage disequilibrium structure of the single nucleotide polymorphism (SNPs) in a reference population to account for the correlation between SNPs. C_LIO_LIIn addition to Mendelian randomization and TWAS, hJAM offers flexibility to incorporate functional or genomic annotation or information from metabolomic studies. C_LI

genetics↗

bioRxiv2019

HiLDA: a statistical approach to investigate differences in mutational signatures

We propose a hierarchical latent Dirichlet allocation model (HiLDA) for characterizing somatic mutation data in cancer. The method allows us to infer mutational patterns and their relative frequencies in a set of tumor mutational catalogs and to compare the estimated frequencies between tumor sets. We apply our method to somatic mutations in colon cancer with mutations classified by the time of occurrence, before or after tumor initiation. Applying the methods to 16 colon cancers, we found significant associations between the relative frequencies of mutational patterns and the time of occurrence of mutations. Our novel method provides higher statistical power for detecting differences in mutational signatures.

genomics↗