Trait-wide Association Results

The UKB has amassed a wealth of health-related information from participants' baseline visits and imaging data at 2nd follow-up visits. Specifically, we incorporated 991 health-related traits, which we organized into 10 distinct categories: mental health (n=235), health and medical history (n=201), diet and food preference (n=174), eye measures (n=109), physical measures (n=95), lifestyle factors (n=77), blood and urine assays (n=54), physical activities (n=29), cognitive function (n=16), and working and living environment (n=1).

Furthermore, our analysis comprised 2,151 imaging traits, spanning various modalities such as brain imaging (including T1 structural brain MRI, susceptibility-weighted brain MRI, and diffusion brain MRI) (n=1978), as well as cardiovascular magnetic resonance (CMR) (n=129) and abdominal MRI traits (n=44). Analyzed traits for detailed UKB field IDs and descriptions.

Trait variables were categorized into three main data types: continuous, ordered categorical, and binary. The assessment of associations between metabolites and traits was conducted using various regression models tailored to the specific types of traits.

Linear regressions were adopted to test continuous and binary traits, while proportional odds logistic regression was applied for ordered categorical traits. Notably, for binary traits in linear regressions, they were treated as independent variables and metabolites as dependent variables. For continuous traits, a pre-processing step of inverse normal transformation (INRT) was employed to ensure data normality. All models adjusted the participants' age, sex, Townsend deprivation index, Body Mass Index (BMI), smoking status, statin medication use, and fasting duration prior to blood collection.

Furthermore, we performed additional sensitivity analysis by adjusting additional covariate of eGFR. Replication analysis in White and non-White ancestries were performed to validate the findings. Results can be accessed through downloaded data files. Detailed data partition criteria and implementation details can be found in the publication at the bottom of the title page.

Analysis were performed by the 'lm' and the 'polr' function from the R package 'MASS' (v4.2.0).