Incorporating prior biological information into genomic predictions: An example from mastitis in Danish Jersey and Nordic Red cattle
Year: 2024
Type: conference-paper
Abstract: Disease risk variants identified through genome-wide association studies (GWAS) are enriched in non-coding genomic regions, suggesting a potential regulatory role.However, regulatory variants identified through expression Quantitative Trait Locus (eQTL) mapping exhibit only a modest overlap with GWAS variants.This may be because traditional eQTL models aggregate RNA profiles from many cells and fail to capture the distinct regulatory programs of causal cell states.We recently proposed a single-cell-resolution eQTL modeling approach (Nathan A et al, 2022) that offers a high-resolution approach to linking genetic variation to gene expression in cell types (such as T cells) or states (e.g.cytotoxicity and proliferation).However, accurately modeling single-cell expression represents a challenge for single-cell eQTL mapping.In some instances, individual genes may exhibit distinct count distribution patterns in single-cell data due to data sparsity and heterogeneous contributions of various cell states.This violates parametric assumptions and may lead to p-value inflation and a higher false-positive rate.Furthermore, current eQTL methods fail to scale to large datasets of millions of cells.Therefore, they are limited to testing only a few variants across a whole locus already defined from pseudobulk analysis.. To solve these problems, we propose a non-parametric linear mixedeffects modeling approach to map cell-state-dependent single-cell eQTLs.Key features of this model include: 1) adaptive bootstrapping to derive statistically calibrated p-values without any assumption about the gene expression distribution, 2) pseudocell aggregation and covariate regression prior to eQTL mapping to reduce computation time while retaining statistical power.We applied this method to 500,089 memory T cells from 259 Peruvian individuals and identified cell-state-dependent eQTLs for autoimmune disease-risk genes such as IL2RA and SH2B3 which had no evidence of eQTL using pseudobulk analysis.Some eQTLs from single-cell resolution analysis were not detected with bulk analysis and had regulatory effects specific to rare states.These results suggest that testing eQTL interactions with cell states can identify cell-state- LIGHT-SPEED WHOLE GENOME ASSOCIATION
Language: English
Subfield: Genetics
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W4412219204