[Preprint] Trait Causal Models: from GWAS, perturbations, and omics to predictive models of human biology, 2026, Dann et al.

Chandelier

Chandelier

Senior Member (Voting Rights)
Trait Causal Models: from GWAS, perturbations, and omics to predictive models of human biology

Dann, Emma; Spence, Jeffrey; Ota, Mineto; Engreitz, Jesse; Marson, Alexander; Pritchard, Jonathan K

Abstract
A central goal of human genetics is to understand the molecular pathways that link genes to functions in cells and tissues, and ultimately to their roles in health and disease. Genome-wide association studies (GWAS) provide a unique — and powerful — tool for finding causal relationships between variants or genes and human traits, yet it remains challenging to gain mechanistic insights from these associations at genome scale.

Here, we argue that recent advances in high-throughput perturbation methods and observational “omics” studies provide valuable complementary data that will be crucial for building the next generation of models in human genetics: formal models linking genes to pathways to cell and tissue biology to traits. These models should be mechanistic, causal, and predictive. We refer to such models as Trait Causal Models.

Web | DOI | Zenodo | Open Access
 
From Bluesky:

Jeff Spence @jeffspence.github.io

It's been 30 years (!!!) since the conceptualization of GWAS.

GWAS has driven tons of discoveries, but with recent experimental developments (CRISPR, single-cell omics), we thought it was time to think about how to drive the next 30 years of human genetics.

Trait Causal Models: from GWAS, perturbations, and omics to predictive models of human biology​

A central goal of human genetics is to understand the molecular pathways that link genes to functions in cells and tissues, and ultimately to their roles in health and disease. Genome-wide association...


We argue that the goal of human genetics should be to build Trait Causal Models. TCMs are quantitative, predictive models that trace the biological processes from genes --> traits.

A graph where nodes are different biological units and edges are the relationships between those units. Effects flow from variants to causal genes to molecular effects like gene programs to cells/tissues to human phenotypes, including clinical phenotypes and diagnoses. Environmental factors also affect some of these nodes.



TCMs should be:

- Quantitative: we should move beyond just cataloging associations and instead we should estimate _by how much_ and _in which direction_ a variant or a perturbation would affect the trait


TCMs should be:

- Mechanistic: a model capturing the mechanisms by which genes affect a trait should be able to convert knowledge about how a perturbation or variant affects an intermediate phenotype into predictions about how it would affect the trait.



To build TCMs will require 3 distinct "data pillars", each with complementary advantages

1. Genetic association data (e.g., GWAS)
2. Perturbations in model systems (e.g., perturb-seq in cell lines)
3. Human subjects omics profiling (e.g., scRNA-seq from cases and controls).

The top shows a trait causal model with effects flowing from causal genes to gene programs to cells/tissues to human phenotypes. Below this are 3 data pillars highlighting which aspects of the trait causal model can be informed by each data pillar. Genome-wide association studies can link variants to genes via eQTLs and genes to human phenotypes, but discoveries are biased due to selection and there is limited power for trans-QTLs. Perturb-seq gives information about the cis- and trans-effects of genetic perturbations, but confound direct and indirect effects, can only be performed in a limited set of accessible cell types, and typically come from small cohorts. Observational omics link molecular phenotypes across cell types and links tissue changes to clinical phenotypes, but these links are non-causal.



There have been incredible advances across all 3 of these data pillars, but we discuss what more needs to be done to move from these pillars to building TCMs.

An overview of how to integrate data pillars to build trait causal models. The first panel shows 2-pillar integrations. GWAS and observational omics can be combined (e.g., cell-type- or program-specific trait heritability enrichment). Omics and Perturb-seq can be combined to find causal pathways / regulators for omics differential expression signatures. Perturb-seq and GWAS can be combined to find gene programs explaining polygenic effects on traits. In he next panel, a prototype trait causal model is shown for red blood cell traits, where genes act through programs including hemoglobin synthesis, cell cycle, and autophagy to affect mean corpuscular hemoglobin. The next panel layers data pillars on a trait causal model showing that perturb-seq relates causal genes to gene programs, GWAS relates causal genes to phenotypes, and omics relates gene programs to trait-associated cell states and phenotypes. The last panel lays out different next-generation trait causal models including multi-cell type models, multi-layer models, and models with feedback.



Key challenges include:

- Extracting directional, quantitative effects of _genes_ on traits from GWAS data
- Choosing cell types and conditions and types of perturbations in screens
- Choosing the right level of abstraction for modeling the path from variant --> trait

The first panel shows how a variant's effect on a trait can be decomposed into beta, its effect on expression, and gamma, the effect of expression on trait. The second panel shows that gamma can be a curve in terms of beta, with different kinds of variants acting on the same gene having different effects on traits. The third panel shows that non-coding variants can be more cell type-specific than coding variants.

The first panel shows how the different data pillars can be layered together to build a trait causal model. First, GWAS data identifies causal genes. Then, perturb-seq relates perturbations of those genes to downstream changes in expression in multiple cell types. Third, observational omics data cna see if those changes in expression correlate with case vs. control status to nominate causal cell types. In the second panel the advantages of omics data are laid out: it is multi-cell type, it can measure associations with the environment, and it can associate changes in clinical biomarkers to changes in gene expression



We discuss all of this and more in the piece.

It was a lot of fun thinking through how to use the data that we have and to try to crystallize what the goal of human genetics should be. We're hoping to spark a conversation so please let us know you're thoughts.

Trait Causal Models: from GWAS, perturbations, and omics to predictive models of human biology​

A central goal of human genetics is to understand the molecular pathways that link genes to functions in cells and tissues, and ultimately to their roles in health and disease. Genome-wide association...


This piece grew out of many conversations about where the field is going, sparked by work from @jengreitz.bsky.social's group, and from @jkpritch.bsky.social's group in collaboration with @marsonlab.bsky.social (led by @minetoota.bsky.social). Thank you all for all of your insights!

And a huge thank you to @emmamarydann.bsky.social for co-leading the effort and to @jkpritch.bsky.social for the vision!

As an addendum, AI slop has made it difficult to post perspective pieces on preprint servers, so we're sharing ours on Zenodo.

We did use LLMs to catch typos etc..., but rest assured that human minds conceived the work, human fingers typed it up, and human eyeballs proofread it.
​
 
Chandelier said:
From Bluesky:

​
​
It's been 30 years (!!!) since the conceptualization of GWAS.​
​
GWAS has driven tons of discoveries, but with recent experimental developments (CRISPR, single-cell omics), we thought it was time to think about how to drive the next 30 years of human genetics.​
​
​

Trait Causal Models: from GWAS, perturbations, and omics to predictive models of human biology​


​
​
We argue that the goal of human genetics should be to build Trait Causal Models. TCMs are quantitative, predictive models that trace the biological processes from genes --> traits.​
​
A graph where nodes are different biological units and edges are the relationships between those units. Effects flow from variants to causal genes to molecular effects like gene programs to cells/tissues to human phenotypes, including clinical phenotypes and diagnoses. Environmental factors also affect some of these nodes.
​
​
TCMs should be:​
​
- Quantitative: we should move beyond just cataloging associations and instead we should estimate _by how much_ and _in which direction_ a variant or a perturbation would affect the trait​
​
​
TCMs should be:​
​
- Mechanistic: a model capturing the mechanisms by which genes affect a trait should be able to convert knowledge about how a perturbation or variant affects an intermediate phenotype into predictions about how it would affect the trait.​
​
​
​
To build TCMs will require 3 distinct "data pillars", each with complementary advantages​
​
1. Genetic association data (e.g., GWAS)​
2. Perturbations in model systems (e.g., perturb-seq in cell lines)​
3. Human subjects omics profiling (e.g., scRNA-seq from cases and controls).​
​
The top shows a trait causal model with effects flowing from causal genes to gene programs to cells/tissues to human phenotypes. Below this are 3 data pillars highlighting which aspects of the trait causal model can be informed by each data pillar. Genome-wide association studies can link variants to genes via eQTLs and genes to human phenotypes, but discoveries are biased due to selection and there is limited power for trans-QTLs. Perturb-seq gives information about the cis- and trans-effects of genetic perturbations, but confound direct and indirect effects, can only be performed in a limited set of accessible cell types, and typically come from small cohorts. Observational omics link molecular phenotypes across cell types and links tissue changes to clinical phenotypes, but these links are non-causal.
​
​
There have been incredible advances across all 3 of these data pillars, but we discuss what more needs to be done to move from these pillars to building TCMs.​
​
An overview of how to integrate data pillars to build trait causal models. The first panel shows 2-pillar integrations. GWAS and observational omics can be combined (e.g., cell-type- or program-specific trait heritability enrichment). Omics and Perturb-seq can be combined to find causal pathways / regulators for omics differential expression signatures. Perturb-seq and GWAS can be combined to find gene programs explaining polygenic effects on traits. In he next panel, a prototype trait causal model is shown for red blood cell traits, where genes act through programs including hemoglobin synthesis, cell cycle, and autophagy to affect mean corpuscular hemoglobin. The next panel layers data pillars on a trait causal model showing that perturb-seq relates causal genes to gene programs, GWAS relates causal genes to phenotypes, and omics relates gene programs to trait-associated cell states and phenotypes. The last panel lays out different next-generation trait causal models including multi-cell type models, multi-layer models, and models with feedback.
​
​
Key challenges include:​
​
- Extracting directional, quantitative effects of _genes_ on traits from GWAS data​
- Choosing cell types and conditions and types of perturbations in screens​
- Choosing the right level of abstraction for modeling the path from variant --> trait​
​
The first panel shows how a variant's effect on a trait can be decomposed into beta, its effect on expression, and gamma, the effect of expression on trait. The second panel shows that gamma can be a curve in terms of beta, with different kinds of variants acting on the same gene having different effects on traits. The third panel shows that non-coding variants can be more cell type-specific than coding variants.'s effect on a trait can be decomposed into beta, its effect on expression, and gamma, the effect of expression on trait. The second panel shows that gamma can be a curve in terms of beta, with different kinds of variants acting on the same gene having different effects on traits. The third panel shows that non-coding variants can be more cell type-specific than coding variants.
The first panel shows how the different data pillars can be layered together to build a trait causal model. First, GWAS data identifies causal genes. Then, perturb-seq relates perturbations of those genes to downstream changes in expression in multiple cell types. Third, observational omics data cna see if those changes in expression correlate with case vs. control status to nominate causal cell types. In the second panel the advantages of omics data are laid out: it is multi-cell type, it can measure associations with the environment, and it can associate changes in clinical biomarkers to changes in gene expression
​
​
We discuss all of this and more in the piece.​
​
It was a lot of fun thinking through how to use the data that we have and to try to crystallize what the goal of human genetics should be. We're hoping to spark a conversation so please let us know you're thoughts.​
​

Trait Causal Models: from GWAS, perturbations, and omics to predictive models of human biology​



​
This piece grew out of many conversations about where the field is going, sparked by work from @jengreitz.bsky.social's group, and from @jkpritch.bsky.social's group in collaboration with @marsonlab.bsky.social (led by @minetoota.bsky.social). Thank you all for all of your insights!​
​
And a huge thank you to @emmamarydann.bsky.social for co-leading the effort and to @jkpritch.bsky.social for the vision!​
​
As an addendum, AI slop has made it difficult to post perspective pieces on preprint servers, so we're sharing ours on Zenodo.​
​
We did use LLMs to catch typos etc..., but rest assured that human minds conceived the work, human fingers typed it up, and human eyeballs proofread it.​
​

Click to expand...
Unfortunate acronym.
 
You must log in or register to reply here.
Back
Top Bottom