Within the Perspective in Nature Genetics titled: The predicament of heritable confounders, collaborators and colleagues of mine elaborate on the effects that deep and shallow phenotyping have on downstream genetic analyses.
I was involved in reviewing and editing of this manuscript, and the perspective uses a finding from my own pre-print to strengthen their own arguements.
Within the perspective, the authors introduce the concept of deep versus shallow phenotyping, a concept already previously highlighted by my PhD supervisor Dr. Na Cai in this paper:
However, the current work takes these findings a step further, and performs simulations to show how the difference in these phenotypes would have an effect on false-positive and false-negative findings. The authors conclude that shallow phenotyping would likely, besides true genetic findings, also find false-positives, and thereby inflate results. This can greatly affect interpretations and assumptions made within downstream analyses of these phenotypes and their GWAS summary statistics. This is also true and extends to real data, and becomes apparent between different cohorts that have used different methods to assess mental health in their participants.
When interpreting results then, from studies based on shallow phenotyping, these issues can create ambiguous genetic correlation estimates and extends to other downstream methods: there is a chance that the latent factor from genomic SEM models, doesn’t capture a latent shared genetic liability, but rather would capture shared bias effects. This was an interpretation of the findings in my first-author pre-print.
The authors then present some solutions to work around the bias and improve our interpretation of methods (directly cited) :
- “all meta-analyses should report between-cohort rGSNP values”
- “metrics of validity beyond rGSNP should be used, especially for analyses focused on polygenic scores (PGS)”
For now the authors emphasize a need to focus on deep phenotyped cohorts over larger sample sizes, such as interview-based and EHR-based datasets. They also suggest to improve phenotypic imputation methods, which could improve power without the need to use high-throughput questionnairs, but instead could work to impute information based on deep-phenotypes.
I am thankfull for the opportunity to colaborrate and be involved in this important work.
Find the full publication here:
Cai, N., Dahl, A., Border, R., Rietkerk,J. et al. The predicament of heritable confounders. Nat Genet (2026). https://doi.org/10.1038/s41588-025-02465-y
Leave a Reply