Genome-wide Causation Studies of Complex Diseases
Pith reviewed 2026-05-24 19:54 UTC · model grok-4.3
The pith
Additive noise models applied to schizophrenia genetics find mostly non-overlapping causation and association signals.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
The paper proposes genome-wide causation studies (GWCS) as an alternative to GWAS and develops additive noise models (ANMs) for genetic causation analysis. Type I error rates and power of the ANMs to test for causation are presented. We conduct GWCS of schizophrenia. Both simulation and real data analysis show that the proportion of the overlapped association and causation signals is small.
What carries the argument
Additive noise models (ANMs) that test for direct causal effects between genetic variants and disease by assuming the effect of one variable on another plus independent noise, allowing distinction from mere statistical association.
If this is right
- GWCS using ANMs can identify causal genetic variants that standard association analysis misses in complex diseases.
- The small overlap indicates that a large fraction of disease-causing variants remain hidden when relying solely on GWAS signals.
- ANMs provide type I error control and power calculations that enable genome-wide testing for causation rather than dependence.
- Shifting focus to causation analysis questions the sufficiency of association as the primary platform for genetic studies of complex diseases.
Where Pith is reading between the lines
- Reanalysis of existing GWAS datasets with ANMs could prioritize different loci for functional follow-up and drug target identification.
- The approach may extend to other complex traits if the additive noise assumption holds across different disease architectures.
- Accounting for linkage disequilibrium within the ANM framework would be a direct next step to increase applicability in real genomes.
Load-bearing premise
Additive noise models can reliably distinguish causation from association in genome-wide genetic data without being invalidated by unmeasured confounders, population structure, or linkage disequilibrium.
What would settle it
Re-running the schizophrenia analysis with ANMs on an independent dataset that includes known causal variants validated by functional experiments or Mendelian randomization and finding substantial overlap between ANM signals and those known causal variants would undermine the claim of small overlap.
read the original abstract
Despite significant progress in dissecting the genetic architecture of complex diseases by genome-wide association studies (GWAS), the signals identified by association analysis may not have specific pathological relevance to diseases so that a large fraction of disease causing genetic variants is still hidden. Association is used to measure dependence between two variables or two sets of variables. Genome-wide association studies test association between a disease and SNPs (or other genetic variants) across the genome. Association analysis may detect superficial patterns between disease and genetic variants. Association signals provide limited information on the causal mechanism of diseases. The use of association analysis as a major analytical platform for genetic studies of complex diseases is a key issue that hampers discovery of the mechanism of diseases, calling into question the ability of GWAS to identify loci underlying diseases. It is time to move beyond association analysis toward techniques enabling the discovery of the underlying causal genetic strctures of complex diseases. To achieve this, we propose a concept of a genome-wide causation studies (GWCS) as an alternative to GWAS and develop additive noise models (ANMs) for genetic causation analysis. Type I error rates and power of the ANMs to test for causation are presented. We conduct GWCS of schizophrenia. Both simulation and real data analysis show that the proportion of the overlapped association and causation signals is small. Thus, we hope that our analysis will stimulate discussion of GWAS and GWCS.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes genome-wide causation studies (GWCS) using additive noise models (ANMs) as an alternative to standard GWAS association analysis for complex diseases. It develops ANMs, reports their type I error rates and power, applies the method to schizophrenia data, and concludes that the proportion of overlapped association and causation signals is small in both simulations and real data.
Significance. If the ANM-based distinction between causation and association holds after proper handling of GWAS-specific artifacts, the work could motivate a shift from association-focused to causation-focused genetic analysis and reduce the fraction of non-causal GWAS hits. The simulation framework and real-data application are potentially valuable if the identifiability assumptions are satisfied.
major comments (3)
- [Abstract and Methods] The central claim that the overlap between association and causation signals is small rests on the ANM correctly identifying causal SNPs. However, ANM identifiability requires that the residual after regressing phenotype on genotype is independent of the genotype; this independence is violated by linkage disequilibrium among SNPs and by population stratification, both ubiquitous in GWAS data. No correction (e.g., principal components, mixed models, or LD pruning) is described.
- [Abstract] The abstract states that type I error rates and power of the ANMs are presented and that quantitative results on the schizophrenia analysis support a small overlap, yet supplies neither the ANM equations, the fitting procedure, nor any numerical results or tables. Without these, the reported small overlap cannot be evaluated or reproduced.
- [Results (schizophrenia analysis)] Standard GWAS practice accounts for population structure precisely because it induces dependence between genotype and phenotype residuals; applying ANM without such adjustment risks attributing stratification-induced dependence to causation, which would artifactually inflate or deflate the reported overlap proportion.
minor comments (1)
- [Abstract] Typo in abstract: 'strctures' should read 'structures'.
Simulated Author's Rebuttal
We thank the referee for the constructive comments. We respond point by point to the major comments, indicating where revisions to the manuscript will be made to address the concerns raised.
read point-by-point responses
-
Referee: [Abstract and Methods] The central claim that the overlap between association and causation signals is small rests on the ANM correctly identifying causal SNPs. However, ANM identifiability requires that the residual after regressing phenotype on genotype is independent of the genotype; this independence is violated by linkage disequilibrium among SNPs and by population stratification, both ubiquitous in GWAS data. No correction (e.g., principal components, mixed models, or LD pruning) is described.
Authors: We agree that ANM identifiability depends on residual-genotype independence and that LD and population stratification violate this in GWAS data. The manuscript introduces the GWCS concept and presents ANM results under the stated assumptions, with simulations assuming independence. The real-data application does not include corrections. We will revise to add explicit discussion of these limitations and recommend preprocessing steps such as LD pruning or principal components before ANM fitting in future applications. revision: yes
-
Referee: [Abstract] The abstract states that type I error rates and power of the ANMs are presented and that quantitative results on the schizophrenia analysis support a small overlap, yet supplies neither the ANM equations, the fitting procedure, nor any numerical results or tables. Without these, the reported small overlap cannot be evaluated or reproduced.
Authors: The abstract is a concise summary. The ANM equations, fitting procedure, type I error and power results, and quantitative schizophrenia findings (including tables) are provided in the Methods and Results sections of the full manuscript. We do not believe the abstract requires expansion with technical details or equations, which are standardly placed in the body. revision: no
-
Referee: [Results (schizophrenia analysis)] Standard GWAS practice accounts for population structure precisely because it induces dependence between genotype and phenotype residuals; applying ANM without such adjustment risks attributing stratification-induced dependence to causation, which would artifactually inflate or deflate the reported overlap proportion.
Authors: We acknowledge this risk. The schizophrenia analysis applied ANM without population-structure adjustment. We will revise the manuscript to discuss this issue explicitly and, where feasible, incorporate adjustments (e.g., principal components) to evaluate the robustness of the reported small overlap. revision: yes
Circularity Check
No circularity: ANM application and overlap statistic are independent of inputs
full rationale
The paper introduces additive noise models (ANMs) for genome-wide causation studies as an alternative to GWAS, presents type I error and power calculations, and reports an empirical finding that the proportion of overlapped association and causation signals is small in both simulations and schizophrenia data. No quoted step shows a self-definitional reduction, a fitted parameter renamed as a prediction, or a load-bearing self-citation chain. The overlap proportion is computed from separate association tests and ANM-based causation tests applied to the same data; nothing in the abstract or described chain indicates that this proportion is forced by construction from the model fitting itself. The derivation chain is therefore self-contained against external benchmarks.
Axiom & Free-Parameter Ledger
axioms (1)
- domain assumption Additive noise models can be used to test for causation between SNPs and disease status across the genome.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.