Causal Inference with Genetic Data: Past, Present, and Future

Using staged tree models for health data: Investigating invasive fungal infections by aspergillus and other filamentous fungi

Machine learning models are increasingly used in the medical domain to study the association between risk factors and diseases to support practitioners in understanding health outcomes. In this paper, we showcase the use of machine-learned staged tree models for investigating complex asymmetric dependence structures in health data. Staged trees are a specific class of generative, probabilistic graphical models that formally model asymmetric conditional independence and non-regular sample spaces. An investigation of the risk factors in invasive fungal infections demonstrates the insights staged trees provide to support medical decision-making.

Causal and common risk pathways linking childhood maltreatment to later intimate partner violence victimization

Childhood maltreatment and intimate partner violence (IPV) victimization are major psychiatric risk factors. Maltreatment substantially increases the likelihood of subsequent IPV victimization, but what drives this association is poorly understood. We analyzed retrospective self-reports of maltreatment and IPV in 12794 participants (58% women, 42% men) from the Twins Early Development Study at ages 21 and 26 using quantitative genetic methods. We estimated the etiological influences common to maltreatment and IPV, and the direct causal effect of maltreatment on IPV beyond such common influences. Participants exposed to maltreatment (~7% of the sample) were 3 times more likely to experience IPV victimization than their peers at age 21, 4 times more likely at 26. The association between maltreatment and IPV was mostly due to environmental influences shared by co-twins (42-43%) and genetic influences (30-33%). The association between maltreatment and IPV was similar for women and men, but its etiology partly differed by sex. Maltreatment had a moderate-to-large effect on IPV in phenotypic models (β = 0.23-0.34), decreasing to a small-to-moderate range in causal models accounting for their common etiology (β = 0.15-0.21). Risk factors common to maltreatment and IPV victimization are largely familial in origin, environmental and genetic. Even considering common risk factors, experiencing maltreatment is causally related to subsequent IPV victimization. Interventions promoting safe intimate relationships among young adults exposed to maltreatment are warranted and should address family-level environmental risk and individual-level risk shaped by genetics.

Reframing sepsis immunobiology for translation: towards informative subtyping and targeted immunomodulatory therapies

Sepsis is a common and deadly condition. Within the current model of sepsis immunobiology, the framing of dysregulated host immune responses into proinflammatory and immunosuppressive responses for the testing of novel treatments has not resulted in successful immunomodulatory therapies. Thus, the recent focus has been to parse observable heterogeneity into subtypes of sepsis to enable personalised immunomodulation. In this Personal View, we highlight that many fundamental immunological concepts such as resistance, disease tolerance, resilience, resolution, and repair are not incorporated into the current sepsis immunobiology model. The focus for addressing heterogeneity in sepsis should be broadened beyond subtyping to encompass the identification of deterministic molecular networks or dominant mechanisms. We explicitly reframe the dysregulated host immune responses in sepsis as altered homoeostasis with pathological disruption of immune-driven resistance, disease tolerance, resilience, and resolution mechanisms. Our proposal highlights opportunities to identify novel treatment targets and could enable successful immunomodulation in the future.

Causal complexity in human research: On the shared challenges of behavior genetics, medical genetics, and environmentally oriented social science

We received 23 spirited commentaries on our target article from across the disciplines of philosophy, economics, evolutionary genetics, molecular biology, criminology, epidemiology, and law. We organize our reply around three overarching questions: (1) What is a cause? (2) How are randomized controlled trials (RCTs) and within-family genome-wide association studies (GWASs) alike and unalike? (3) Is behavior genetics a qualitatively different enterprise? Throughout our discussion of these questions, we advocate for the idea that behavior genetics shares many of the same pitfalls and promises as environmentally oriented research, medical genetics, and other arenas of the social and behavioral sciences.

The providential randomisation of genotypes

When building causal knowledge in behavioural genetics, the natural randomisation of genotypes at conception (approximately analogous to the artificial randomisation occurring in randomised controlled trials) facilitates the discovery of genetic causes. More importantly, the randomisation of genetic material within families also enables a better identification of (environmental) risk factors and aetiological pathways to diseases and behaviours.

mGWAS-Explorer 2.0: Causal Analysis and Interpretation of Metabolite-Phenotype Associations

Metabolomics-based genome-wide association studies (mGWAS) are key to understanding the genetic regulations of metabolites in complex phenotypes. We previously developed mGWAS-Explorer 1.0 to link single-nucleotide polymorphisms (SNPs), metabolites, genes and phenotypes for hypothesis generation. It has become clear that identifying potential causal relationships between metabolites and phenotypes, as well as providing deep functional insights, are crucial for further downstream applications. Here, we introduce mGWAS-Explorer 2.0 to support the causal analysis between >4000 metabolites and various phenotypes. The results can be interpreted within the context of semantic triples and molecular quantitative trait loci (QTL) data. The underlying R package is released for reproducible analysis. Using two case studies, we demonstrate that mGWAS-Explorer 2.0 is able to detect potential causal relationships between arachidonic acid and Crohn's disease, as well as between glycine and coronary heart disease.

Genetic nurture versus genetic transmission of risk for ADHD traits in the Norwegian Mother, Father and Child Cohort Study

Identifying mechanisms underlying the intergenerational transmission of risk for attention-deficit/hyperactivity disorder (ADHD) traits can inform interventions and provide insights into the role of parents in shaping their children's outcomes. We investigated whether genetic transmission and genetic nurture (environmentally mediated effects) underlie associations between polygenic scores indexing parental risk and protective factors and their offspring's ADHD traits. This birth cohort study included 19,506 genotyped mother-father-offspring trios from the Norwegian Mother, Father and Child Cohort Study. Polygenic scores were calculated for parental factors previously associated with ADHD, including psychopathology, substance use, neuroticism, educational attainment, and cognitive performance. Mothers reported on their 8-year-old children's ADHD traits (n = 9,454 children) using the Parent/Teacher Rating Scale for Disruptive Behaviour Disorders. We found that associations between ADHD maternal and paternal polygenic scores and child ADHD traits decreased significantly when adjusting for the child polygenic score (pΔβ = 9.95 × 10-17 for maternal and pΔβ = 1.48 × 10-14 for paternal estimates), suggesting genetic transmission of ADHD risk. Similar patterns suggesting genetic transmission of risk were observed for smoking, educational attainment, and cognition. The maternal polygenic score for neuroticism remained associated with children's ADHD ratings even after adjusting for the child polygenic score, indicating genetic nurture. There was no robust evidence of genetic nurture for other parental factors. Our findings indicate that the intergenerational transmission of risk for ADHD traits is largely explained by the transmission of genetic variants from parents to offspring rather than by genetic nurture. Observational associations between parental factors and childhood ADHD outcomes should not be interpreted as evidence for predominantly environmentally mediated effects.

Research Review: How to interpret associations between polygenic scores, environmental risks, and phenotypes

BACKGROUND

Genetic influences are ubiquitous as virtually all phenotypes and most exposures typically classified as environmental have been found to be heritable. A polygenic score summarises the associations between millions of genetic variants and an outcome in a single value for each individual. Ever lowering costs have enabled the genotyping of many samples relevant to child psychology and psychiatry research, including cohort studies, leading to the proliferation of polygenic score studies. It is tempting to assume that associations detected between polygenic scores and phenotypes in those studies only reflect genetic effects. However, such associations can reflect many pathways (e.g. via environmental mediation) and biases.

METHODS

Here, we provide a comprehensive overview of the many reasons why associations between polygenic scores, environmental exposures, and phenotypes exist. We include formal representations of common analyses in polygenic score studies using structural equation modelling. We derive biases, provide illustrative empirical examples and, when possible, mention steps that can be taken to alleviate those biases.

RESULTS

Structural equation models and derivations show the many complexities arising from jointly modelling polygenic scores with environmental exposures and phenotypes. Counter-intuitive examples include that: (a) associations between polygenic scores and phenotypes may exist even in the absence of direct genetic effects; (b) associations between child polygenic scores and environmental exposures can exist in the absence of evocative/active gene-environment correlations; and (c) adjusting an exposure-outcome association for a polygenic score can increase rather than decrease bias.

CONCLUSIONS

Strikingly, using polygenic scores may, in some cases, lead to more bias than not using them. Appropriately conducting and interpreting polygenic score studies thus requires researchers in child psychology and psychiatry and beyond to be versed in both epidemiological and genetic methods or build on interdisciplinary collaborations.