Florian Privé

This study assesses the progress and limitations of polygenic risk score prediction accuracy for common diseases, showing diminishing returns from recent GWAS and highlighting the need for broader variant coverage.

This study examines the genetic and phenotypic links between neonatal C3 and C4 protein concentrations and psychiatric and autoimmune disorders, finding no strong causal associations with mental disorders but evidence of lower autoimmune risk with higher C4 levels.

This paper validates LDpred2-auto for inferring genetic parameters and shows improved heritability capture and predictive performance with an expanded variant set.

ADuLT is a time-to-event GWAS method that outperforms Cox regression in strongly ascertained studies, finding more genome-wide significant associations in psychiatric disorder analyses.

This study identifies genetic loci associated with vitamin D-binding protein and 25 hydroxyvitamin D in neonates and links DBP to vitamin D status and disease risk.

The paper shows that standard methods for liability-scale heritability estimation in GWAS meta-analysis can be downwardly biased and proposes a simple correction based on effective sample size summation.

This paper presents a publicly available privacy-preserving benchmark for polygenic prediction using summary-level data, enabling accurate model performance estimation without access to individual-level genetic data.

The paper evaluates polygenic score accuracy across genetic ancestry continua in diverse populations, finding significant variation in performance linked to genetic distance from training data.

This paper introduces a framework to enhance polygenic prediction by combining multiple polygenic scores, showing significant accuracy improvements for psychiatric disorders and other traits.

The paper identifies misspecifications in GWAS summary statistics and shows how to correct them to improve polygenic score predictive performance.

The paper defines 21 ancestry groups from UK Biobank data and develops a method to measure ancestry proportions from GWAS summary statistics using allele frequencies and principal components.

This paper introduces LT-FH++, a method that improves GWAS power by incorporating age-of-onset and family history, showing increased significant associations in psychiatric disorders and mortality data.

The paper develops an optimal algorithm for splitting the genome into linkage disequilibrium blocks using dynamic programming and implements it in the bigsnpr R package.

This paper develops and compares genetic prediction tools that incorporate heritability models, showing improved accuracy for complex traits using individual-level and summary statistics data.

This study compares methods for combining GWAS summary statistics and individual-level data to improve polygenic prediction, finding that Meta-PRS is often more accurate than Meta-GWAS.

This paper introduces an analytical method to estimate effective sample size in mixed-model GWAS of quantitative traits, showing its performance in unrelated and related individuals.

LDpred2 reworks LDpred with a 3 cM LD window plus sparse and auto options, reaching mean AUC 65.1% over 8 real traits versus 63.8% for lassosum.

The paper shows that individual PRS estimates have large uncertainty, impacting risk stratification, and introduces a method to quantify this uncertainty using Bayesian approaches.

This paper benchmarks the performance of two R packages, snpnet and bigstatsr, for fitting penalized regressions on large genetic data, finding bigstatsr significantly faster.

The paper presents best practices and efficient implementations for PCA of genetic data to address issues like LD structure capture, shrinkage bias, and population structure confounding.

This study evaluates DNA methylation deconvolution methods for cell-type proportion inference, identifies critical steps for accuracy, and develops a benchmark pipeline.

The paper improves polygenic score prediction by optimizing multiple hyperparameters in Clumping and Thresholding and introduces Stacked Clumping and Thresholding (SCT) for better performance.

This paper introduces an efficient penalized logistic regression method for genetic risk prediction, showing improved performance over existing methods like C+T and random forests.

VarExp estimates phenotypic variance explained by genetic variants using GWAS summary statistics without individual-level data.

The paper introduces two R packages, bigstatsr and bigsnpr, for efficiently managing and analyzing large-scale genomic data within a single framework.