Gene-Level Pleiotropy Score (gPS)
Summary¶
The gene-level pleiotropy score (gPS) is the count of unique diseases associated with all GWAS credible-set variants that share a likely causal gene. It is a deliberately simple aggregate designed to be computable genome-wide, and its authors are explicit that it measures apparent pleiotropy — an upper bound, not a clean estimate of horizontal pleiotropy.[1]
Source Status¶
Preprint: defined in a bioRxiv preprint not yet peer reviewed.
confidence: medium.
Definition¶
For a gene $g$ with likely-causal assignment to a set of credible sets $C_g$:
$$\text{gPS}(g) = \left| \bigcup_{c \in C_g} \text{diseases}(c) \right|$$
The companion variant pleiotropy score (vPS) applies the same count at the variant rather than gene level.[1]
Gene-to-credible-set assignment in the source study used a revised Locus-to-Gene (L2G) machine learning model, supported by colocalisation against molecular QTL credible sets across 98 tissues or cell types.[1]
Interpretation and caveats¶
These are stated by the authors and matter for any reuse:[1]
- gPS is an upper bound on pleiotropy. It combines horizontal and vertical pleiotropy and is inflated by genetic correlation among the diseases counted.
- A lower bound is available: the number of unique therapeutic areas (TAs) per gene, assuming TA independence. gPS and TA count correlate strongly (Spearman ρ = 0.92), so either can be used, but they bracket the true value rather than agreeing on it.
- gPS is strongly confounded by statistical power. The number of variants assigned to a gene predicts gPS at Spearman ρ = 0.87, and maximum GWAS sample size is the single strongest covariate (univariate β = 1.82, P < 10⁻³⁰⁸).
- gPS grows over time by construction. Retrospective analysis shows exponential growth in gPS as GWAS accumulate — a gene's score is a statement about the current evidence base, not a fixed biological property. Record the data freeze when quoting a value.
- Most of the variance is unexplained. A joint model of 9 covariates explained only 15% of gPS variance (Pearson R² = 0.15).
Reference values¶
From 100,526 GWAS yielding 789,453 credible sets (Open Targets Gentropy, 2026 data freeze):[1]
| Quantity | Value |
|---|---|
| Disease-associated genes | 8,285 |
| Genes with gPS > 1 | 5,314 (64%), mean gPS 4.45, max 148 |
| Genes linked to >1 therapeutic area | 4,743 (57%), mean 2.53, max 21 |
| Highest-gPS genes | CDKN2B (148, across 21 TAs), FTO (126), APOE (107), ABO (105) |
Associations¶
Higher gPS is positively associated with loss-of-function constraint (β = 0.64), number of pathways a gene participates in (β = 3.45), gene length (β = 2.17), and the presence of protein-altering variant associations (β = 0.53); and negatively associated with tissue specificity of the transcriptional profile (β = −0.11).[1]
See Also¶
- Gene-Level Pleiotropy and Therapeutic Safety — how gPS is used to stratify drug-target risk.
- Open Targets — the platform whose Gentropy framework computes it.
Citations¶
[1] Tsepilov, Y. A., Suveges, D., Considine, D. et al. (2026). The Human Pleiotropic Map of GWAS Associations and Therapeutic Implications. bioRxiv (preprint, posted 1 May 2026). Supports: the gPS and vPS definitions; all stated caveats; the reference value table; the covariate associations. Location: Results ("Panoramic view across 100,526 complex trait GWAS"; "Gene pleiotropy reflects functional specialisation and organism-level essentiality"); Fig. 4b. Source paper: 2026.04.28.721048v1.full.pdf