PWAS Hub: exploring gene-based associations of complex diseases with sex dependency

Abstract

The Proteome-Wide Association Study (PWAS) is a protein-based genetic association approach designed to complement traditional variant-based methods like GWAS. PWAS operates in two stages: first, machine learning models predict the impact of genetic variants on protein-coding genes, generating effect scores. These scores are then aggregated into a gene-damaging score for each individual. This score is then used in case-control statistical tests to significantly link to specific phenotypes. PWAS Hub (v1.2) is a user-friendly platform that facilitates the exploration of gene-disease associations using clinical and genetic data from the UK Biobank (UKB), encompassing 500k individuals. PWAS Hub reports on 819 diseases and phenotypes determined by PheCode and ICD-10 clinical codes, each with a minimum of 400 affected individuals. PWAS-derived gene associations were reported for 72% of the tested phenotypes. The PWAS Hub also analyzes gene associations separately for males and females, considering sex-specific genetic effects, inheritance patterns (dominant and recessive), and gene pleiotropy. We illustrated the utility of the PWAS Hub for primary (essential) hypertension (I10), type 2 diabetes mellitus (E11), and specified haematuria (R31) that showed sex-dependent genetic signals. The PWAS Hub, available at pwas.huji.ac.il, is a valuable resource for studying genetic contributions to common diseases and sex-specific effects.

Graphical Abstract

Figure 1.

Searching and browsing options of PWAS Hub database. (A) Multiple actions that can be performed in the PWAS Hub (‘What can I do here?’ panel) with a brief overview of the portal utility, a link to a detailed tutorial and several preprocessed examples. (B) Gene search page with 4 visualization panels (from top to bottom): (i) List identifies genes by the partial text or gene ID; (ii) Gene information with cross references and genomic location; (iii) Variant location on the gene structure with its exon organization and Pfam/InterPro domains; (iv) Gene-centric phenotype table that indicate the associations and statistics for all studied phenotypes. (C) Phenotype browsing (A to Z). The disease / phenotype description is attached to its unique ICD-10 code/ Each phenotype is shown with color coded for ICD-10 I10, Essential (primary) hypertension. (D) Top genes for I10 with statistical values for dominant, recessive and hybrid gene heritability models. Genes are sorted by the q-values (based on hybrid model) and colored by the effect size as a gradient with blue and red colors indicating reduced and increased risk, respectively.

Figure 2.

Gene-phenotype analysis page for the top genes of T2D (ICD-10 E11). The colored rectangles (right) are colored for the type of information that is included. The retrieved results (yellow) indicating the number of results by the symbols F, D, R and C (for FDR, dominant, recessive and hybrid inheritance). The summary header also allows a choice to visualize only significant genes or predetermined gene lists with different number of associated genes (25, 50, 100 or 200). The associated gene statistics (light green) explain the rank according to the q-values. In addition, the gene effect size is determined by Cohen's d statistics. The genes are colored by a combination of the effect size and the q-value. Blueish and reddish colors assigned with a decrease and increased risk for T2D. Genes can be also sorted by the p-value according to underlying inheritance model of dominant, recessive or hybrid. The table support further navigation (light blue) for G and A to the ‘Gene’ page and the ‘Association’ page, respectively.

Figure 3.

Association (of gene-to-phenotype) analysis page for the top genes of ICD-10 R31 (Unspecified haematuria), by sex. The color depicts the strength of the effect and its directionality versus the phenotype. For illustration purposes, only the top 7 genes are shown for both sexes. The male and female groups are indicated by the green icon (left of each sub-table). For a detailed description of the table headers and coloring, see Figure 2.

Figure 4.

Sex-comparison page results and graphical panels for the association of ICD-10 R31. (A) Entry point to the Sex-comparison page and summary statistics of significant genes by sex groups and the total number of genes within all groups. (B) Venn diagram with the numbers of overlapping genes and their identities. (C) The statistical significance of the combination of sex groups relative to the exclusive subsets males, females and both (shown in the gray hatching). (D) Two-sided Manhattan ‘mirror’ plot for coding GWAS (cGWAS) results. The bottom brush widget displays a map of the autosomal chromosomes (1–22) followed by the X and Y chromosomes. Blue bars in the brush tool display locations of significant genes where one could zoom in and thus change the viewport. The y-axis gives significance values to genes for which the x-axis provides the chromosomal coordinates. Dashed horizontal lines mark a p-value of 2e-05 as the significance threshold. The bars extending upward (in blue) are gene significance values from the male group, and the ones extending downward (in red) are for the female group. Entering a zoom level similar to the image shown here exposes the gene names as well. A popup box (yellow frame) with further information on each of the displayed genes is visible when hovering over each bar. The position of the MHC locus is indicated by the horizontal green arrow in (D). Last, panel (E) shows a ranked list of genes by their statistics, and chromosomal position. The analyzed group is marked as F or M, for females and males, respectively.

Figure 5.

Association page and a gene-centric analysis. (A) Entry point for the top associated genes PPP1R18 gene for female group with the phenotype R31. Advanced analysis is provided by interactive tabs (Extended summary and Risk score). (B) Gene view with its 3 exons and 38 ordered variants. The numbered lollipops represent the coding variants that are included in the PWAS analysis, numbered along the protein length. Each variant is linked to dbSNP database through a popup box. The significant variants (P-values < 2e-05) are colored red and marked by a square. (C) All gene variants are sorted by their association P-value and the log odds of the variant is presented. Only P-value <0.05 are colored. (D, E) Population partition of the PPP1R18 genes for dominant (light red, top) and recessive (light green, bottom) models. The gene is protective as shown by the ‘hand’ icon. Horizontal dashed line indicates the average prevalent (in %) of R31 in the studied group. A table with the numbers of cases and controls and partition of the population by the gene effects are displayed. The compared statistical (Fisher's exact test) is reported and the significant P-value is <10^–4.