iiinova — where your digital self keeps evolving

Happy to receive any feedback you may have! Very grateful to everyone involved in this work, huge thanks to Simon Wiegrebe, Thomas Winkler (@winkusch.bsky.social) & Zoltán Kutalik (@zkutalik.bsky.social ) 👏

000

9/ Finally, we examined the role of non-linear age-varying genetic effects. While such effects could contribute to discrepancies between the two designs, given the differing age ranges in cross-sectional and longitudinal samples, they explained little of the observed differences.

100

8/ Focusing on other factors, we found that selective participation also contributed to differences between the two designs. This may reflect distinct participation mechanisms, such as selective enrolment in cross-sectional samples versus survival/dropout in longitudinal samples.

7/ However, effect size estimates showed less agreement between the two designs (r = 0.74). Similar to the phenotypic findings, differences were primarily due to gene-by-cohort effects, where genetic associations vary across birth years, introducing bias into cross-sectional estimates.

6/ Among the identified SNPs, 86% showed consistent interpretation across designs regarding the direction of age-varying genetic effects. These included both attenuation with age (e.g., for obesogenic traits) and intensification over time (e.g., for disease burden and medication use).

5/ At the genetic level, we identified 57 SNPs with significant age-varying effects. Most were detected in the cross-sectional design, likely reflecting greater statistical power due to larger sample sizes and broader age ranges.

100

4/ We observed that this likely reflects confounding by year-of-birth effects (e.g., younger cohorts tend to smoke less), which can bias age estimates in cross-sectional analyses.

3/ RESULTS: At the phenotypic level, cross-sectional and longitudinal age effects showed only moderate agreement. For several traits, especially lifestyle behaviours, effects differed in their direction: e.g., smoking appeared to increase with age cross-sectionally but declined longitudinally.

2/ Using data on 31 health-related traits from the UK Biobank, we focused on two questions: 🔹 Do the two designs lead to the same conclusions? 🔹 If not, what are the sources of bias that account for the observed discrepancies?

1/ We compare two common approaches to modeling age-varying genetic effects: 🔹 Cross-sectional: Comparing genetic associations across individuals of different ages. 🔹 Longitudinal: Estimating genetic effects on change over time within the same individuals.

100

Design and model choices shape inference of age-varying genetic effects on complex traits

🚨New preprint is out! How do genetic effects on complex traits change with age? In this work, we compare different approaches to obtain age-varying genetic effects, and show how design and modeling choices can impact the conclusions we draw. shorturl.at/17snd A thread 🧵👇

shorturl.at

Understanding how genetic influences on complex traits change with age is a fundamental question in genetic epidemiology. Both cross-sectional (between-subject) and longitudinal (within-subject) appro...

14219

Reposted by Tabea Schoeler

Adriaan @adriaan-vd-graaf.bsky.social · 04/07/2025

Our paper, MR-link-2 has just been published! Offering pleiotropy robust Mendelian randomization from a single region! www.nature.com/articles/s41...

A network of metabolites and their potential causal relationships. Green edges are Detected by the statistical causal inference method MR-link-2

23415

Reposted by Tabea Schoeler

Swiss Network for Dementia Research @handle.invalid · 24/06/2025

READING RECOMMENDATION 📖 Genes, environment, and aging: What determines cognitive and physical decline? Using newly available longitudinal aging data from the UK Biobank, @tabeasch.bsky.social and colleagues uncovered genetic and environmental factors influencing cognitive and physical decline.

nature.com

Combining cross-sectional and longitudinal genomic approaches to identify determinants of cognitive and physical decline - Nature Communications

The genetics of human aging remains poorly understood due to limited longitudinal data. Here, the authors show distinct genetic architectures for baseline function and decline in cognitive and physica...

021

Reposted by Tabea Schoeler

Eivind Ystrom @eivindy.bsky.social · 13/06/2025

🧵 THREAD: Preprint from @chrisrayner.bsky.social reveals how to fix selection bias in the Norwegian #MoBa study using population-wide registry data! For the first time, we can quantify and adjust for selection bias in this major epidemiological resource. Spread the good news! osf.io/preprints/os...

osf.io

OSF

22310

Reposted by Tabea Schoeler

Margherita Malanchini @marghmalanchini.bsky.social · 10/06/2025

🧠🧬🧑‍🤝‍🧑 New CoDE Lab study: Disorder-specific genetic effects drive the associations between psychopathology and cognitive functioning. Link to preprint: www.medrxiv.org/content/10.1... Led by the brilliant Wangjingyi Liao 🌟 A short thread summarising the study👇

165

Reposted by Tabea Schoeler

Ted Schwaba @tedmond.bsky.social · 20/05/2025

Extremely excited to share the first effort of the Revived Genomics of Personality Consortium: A highly-powered, comprehensive GWAS of the Big Five personality traits in 1.14 million participants from 46 cohorts. www.biorxiv.org/content/10.1...

515370

Reposted by Tabea Schoeler

Aysu Okbay @aysuo.bsky.social · 20/05/2025

PGI Repository v2.0 preprint out! A 🧵 on the main results and updates @robel-alemu.bsky.social @paturley.bsky.social @alextisyoung.bsky.social

13615

All GWA summary statistics will be soon available @gwascatalog.bsky.social (accession codes GCST90565836-GCST90565865)! As always, wonderful teamwork with @zkutalik.bsky.social‪ and Jean-Baptiste Pingault @atcmap.bsky.social 🙂

121

Further, we found little evidence of common risks shared by (cross-sectional) level of functioning and (longitudinal) decline in cognitive and physical outcomes (11/11)

100

Using Mendelian Randomization to identify risk factors involved in age-related decline, we found that most risks were specific to either cognitive decline (e.g., Alzheimer’s disease liability) or physical decline (e.g., shorter telomere length, higher bone mineral density) (10/)

100

In line with the simulation results, longitudinal genetic effects obtained from baseline-adjusted change falsely captured substantial parts of the baseline genetic effects. Our recommendations are therefore in line with previous discussions in discouraging the use of baseline-adjusted change (9/)

100

In total, 7 loci associated with longitudinal decline, implicating APOE and DUSP6 as the top genes associated with cognitive and physical decline, respectively. Overall, there was little overlap between the genetics of decline and physical/cognitive function (8/)

100

We found that SNP heritability of decline was low (e.g., 3% for physical decline), substantially lower than the heritability observed for levels of function (e.g., 31% for physical function) (7/)

100

We then performed genome-wide tests on the two dimensions of decline, and compared the results to GWA performed on their cross-sectional counterparts (i.e., cognitive/physical function at baseline) (6/)

100

For our main phenotypic outcomes, we curated two composite scores capturing overall cognitive decline (based on 5 cognitive traits, such as fluid intelligence, reaction time) and overall physical decline (including 3 physical traits, such as grip strength, FEV) (5/)

100

We find that residual change scores introduce bias by falsely associating cross-sectional (time-invariant) genetic effects with change. In contrast, models capturing absolute and relative change are more robust in identifying longitudinal genetic effects (4/)

100

We focused on three definitions of change, including (1) difference scores (difference between the follow-up and the baseline phenotype), (2) residual change scores (baseline-adjusted change) and (3) log-difference scores (relative change) (3/)

100

We first performed simulations to identify which definition of change is most appropriate for the discovery of longitudinal genetic effects when using two-wave data (i.e., baseline and one follow-up in the UKB) (2/)

100