Fig. 1

Illustrative study overview. a Summary of the UK Biobank genotype and phenotype data used in the study. We included White British individuals and analyzed LD-pruned and quality-controlled variants in relation to 2,138 phenotypes with a minimum of 100 individuals as cases (binary phenotypes) or non-missing values (quantitative phenotypes) (Supplementary Table 1, Supplementary Data 1). b Truncated singular value decomposition (TSVD) applied to decompose genome- and-phenome-wide summary statistic matrix W to characterize latent components. U, S, and V represent resulting matrices of singular values and vectors. c Decomposition of Genetic Associations (DeGAs) characterizes latent genetic components, which are represented as different colors on the palette, with an unsupervised learning approach — TSVD, followed by identification of the key components for each phenotype of our interest (painting phenotypes with colors) and annotation of each of the components with driving phenotypes, variants, and genes (finding the meanings of colors). d We used the squared cosine score and the contribution score, to quantify compositions and biomedical relevance of latent components. e We applied the genomic region enrichment analysis tool (GREAT) for biological characterization of each component and performed functional experiments focusing on adipocyte biology