Multi-ancestry modeling improves fine-mapping resolution, protein prediction, and discovery for proteome-wide...

Area of Science:

• Genomics
• Proteomics
• Population Genetics

Background:

• Proteomic predictive models often lack accuracy in diverse populations due to training on European-ancestry data.
• Limited power and generalizability of current models hinder their application in ancestrally diverse groups.

Purpose of the Study:

• To develop and evaluate proteomic prediction models using whole-genome sequencing and plasma protein data across multiple ancestries.
• To improve fine-mapping resolution and protein-phenotype association discovery in diverse populations.

Main Methods:

• Performed cis- and trans-protein quantitative trait locus (pQTL) mapping across European, African, Hispanic, and Chinese ancestry groups.
• Benchmarked fine-mapping models (SuSiE, SuShiE, MultiSuSiE, SuSiEx) and protein-prediction models (MASHR, UDR, EN) using multi-ancestry cohorts.
• Applied developed models in proteome-wide association studies (PWAS) for 10 phenotypes.

Main Results:

• African-ancestry samples showed improved fine-mapping resolution and smaller cis-credible sets due to shorter linkage disequilibrium (LD) and greater allele frequency diversity.
• Multivariate adaptive shrinkage (MASHR) and ultimate deconvolution in R (UDR) models outperformed elastic net (EN) regression for protein prediction.
• Inclusion of trans-pQTLs and fine-mapping improved prediction performance and PWAS discovery, identifying 68 novel protein-phenotype associations.

Conclusions:

• Including multiple ancestries in genomic studies is crucial for capturing regulatory variation and improving cross-ancestry generalizability of proteomic models.
• Advanced models like MASHR and UDR significantly increase protein-phenotype association discovery compared to traditional methods.
• The study highlights the need for diverse genomic data to enhance precision medicine applications.