Multi-Scale GWAS Translation

Summary

Two decades after the first GWAS, the field's centre of gravity has moved from locus discovery to functional interpretation across scales — proteins, cells, organs. A 2026 review from the Inouye group organises this translation into three pillars: therapeutic target prioritization, cellular architectures of disease, and imaging genetics linking variants to organ structure and function.[1] This page records that framing and, more usefully, the methodological limits the review names for each pillar.

Source Status

Source type: this page is compiled from a review, and per §3.3A reviews are used here for framing and consensus rather than as the source of any precise result. Claims attributed to individual primary studies below are attributed as the review reports them and were not independently verified against those originals. confidence: medium.

The three pillars

Pillar I — Genetics for therapeutic target prioritization

Intersecting GWAS signals with druggable gene databases to prioritise targets, extended to drug repurposing and pharmacogenomics.[1] This vault holds primary-source coverage of the underlying evidence: see Genetics-Guided Therapeutic Target Discovery, and Gene-Level Pleiotropy and Therapeutic Safety for the quantitative treatment of the safety side of the same problem.

Pillar II — Cellular architectures of disease

Two method families:[1]

1. Heritability enrichment, built on stratified LD score regression (S-LDSC). Rather than mapping individual variants to cells, these methods test whether trait heritability concentrates in genomic regions defined by cell-type-specific expression or chromatin accessibility. Extensions named include CELLECT (multiple expression-specificity metrics from scRNA-seq), sc-linker (continuous annotations plus tissue-specific enhancer–gene linking), S-LDSC combined with scATAC-seq, spatial-transcriptomic annotations revealing heritability gradients within brain and liver, and scDRS.
2. Single-cell QTLs and causal inference of cell types, including genotype-neighborhood associations (Rumker et al.) that identify variants influencing the abundance of cell states rather than working within predefined cell types.

Reported applications include a hypothalamus spatio-cellular map showing human energy balance is neuron-centric and regionally organised (recovering MC4R, PCSK1, POMC, CALCR plus BSN and CORO1A); etiological heterogeneity in type 2 diabetes across pancreatic islets, adipocytes, endothelial and enteroendocrine cells; coronary artery disease risk mediated predominantly through smooth muscle and endothelial cells; and site-specific cell-type differences in osteoarthritis, where interzone chondrocytes contribute specifically to total hip replacement risk.[1]

The stated limitation is the important part. Heritability-enrichment methods require high polygenicity for stable estimates, and their power is sensitive to annotation size — they are primarily used with annotations covering >0.5% of SNPs. When the genomic footprint of linked enhancers or cell-type-specific peaks is too small, heritability estimates become unstable and type I error inflates.[1] This is a hard constraint on how finely cell types can be resolved, independent of single-cell data quality.

Pillar III — Imaging genetics

GWAS of imaging-derived organ structure and function traits, aggregated into polygenic scores and used for risk prediction and Mendelian randomization. Reported examples: a PGS for left atrial passive emptying fraction predictive of ischemic stroke; right-ventricular trait PGSs associated with coronary artery disease and dilated cardiomyopathy; left-ventricular trait PGSs predictive of heart failure and dilated cardiomyopathy; and an aortic-diameter PGS improving prediction of aortic dilation and adverse thoracic aortic events beyond clinical risk factors.[1] The review also covers multi-organ connections and moving beyond human-defined imaging phenotypes toward learned representations.

Why the three-pillar framing is useful

Lab interpretation: the pillars are not independent lines of work — they are three different answers to the same question ("what does this locus do?"), asked at different scales, and each has a characteristic blind spot. Target prioritization tells you which gene but not in which cell or tissue. Cell-type enrichment tells you where, but only for highly polygenic traits with large enough annotations [1]. Imaging genetics gives an organ-level phenotype that is closer to clinical endpoints but further from molecular mechanism. A locus that resists interpretation in one pillar is often tractable in another, which is an argument for running them in parallel rather than sequentially. The review presents the pillars descriptively; treating them as complementary failure modes is this vault's reading.

This framing also connects directly to work already held here: Tissue-Partitioned Heritability is the Pillar II machinery, Cell-Type Interaction QTLs an adjacent route to the same target, and Genetics of Subclinical Coronary Atherosclerosis Imaging plus Oculomics are Pillar III instances.

See Also

• Genetics-Guided Therapeutic Target Discovery — Pillar I, with primary sourcing.
• Gene-Level Pleiotropy and Therapeutic Safety — the safety dimension of Pillar I.
• Tissue-Partitioned Heritability — the S-LDSC family underpinning Pillar II.
• Cell-Type Interaction QTLs — cell-context QTL mapping.
• Genetics of Subclinical Coronary Atherosclerosis Imaging, Oculomics — Pillar III instances.
• Sequence-to-Function Genomic Models — the molecular-scale route the review's pillars sit above.

Citations

[1] Felici, B., Chen, S., Yuan, M., Jiang, X., Ip, S., Rudd, J. H. F. & Inouye, M. (2026). Translating genome-wide association studies at multiple scales: Drug target prioritization, cellular architectures, and organ imaging. Cell Genomics 6:101282 (Review). Supports: the three-pillar framing; the enumeration of Pillar II methods and their reported applications; the polygenicity and annotation-size limitations of heritability enrichment; the imaging-PGS examples in Pillar III. Location: Summary; Introduction; "Pillar I: Genetics for therapeutic target prioritization and drug discovery"; "Pillar II: Cellular architectures of disease" (including "Prioritizing disease-relevant cell types via heritability enrichment" and "Single-cell QTLs and causal inference of cell types"); "Pillar III: Imaging connects genetic variants to organ-level physiology and function traits" (including "Polygenic scores of imaging traits"). Source paper: PIIS2666979X26001448.pdf