Genetic association data from genome-wide association studies (GWAS) are foundational for our understanding of complex diseases and traits. But in order to apply these results to diagnosis, drug development, and treatment, we need to identify the effector genes that explain those genetic associations. This is rarely straightforward: most SNPs associated with disease are located outside of coding regions of the genome, so that their impact on genes is not obvious; and even a variant located in a protein-coding gene may actually affect a different gene. And to complicate things further, a variant that is strongly associated with disease may not have a direct impact on a gene, but may rather be "along for the ride" with a tightly linked causal variant.
To help bridge the gap between genetic association results and the effector genes that are directly involved in disease, we are aggregating additional data types—for example, transcriptional regulation, tissue specificity, curated biological annotations, and more—and integrating them, using cutting-edge computational methods, in order to mine insights from GWAS data. We present the results of these methods in interactive FOCUS (Find Orthogonal Computational Support) tables.
As a first step in implementing these methods, we needed to find a way to store many different connections between variants, genes, tissues, phenotypes, and biological annotations. We decided to use a Neo4J graph database, which holds data nodes and their relationships with each other and can support complex, scientifically meaningful queries.
|Neo4J graph showing variants on chromosome 8 that are associated with glycemic phenotypes. Orange circles represent variants; pink, p-values; blue, phenotypes; red, phenotype group; green and brown, variant annotations.|
We have also created pipelines to apply computational methods to the genetic association data in the Knowledge Portal Network. In brief, we are currently running:
- MetaXcan, which integrates tissue-specific expression data from GTEx and genetic association data to predict the potential that a gene is causal for a phenotype in a given tissue;
- DEPICT, which integrates multiple data sources including transcriptional co-regulation, Gene Ontology annotations, model organism phenotypes, and more to make several predictions: membership of a gene in a pathway; the probability of its association with a given phenotype; and the tissues or cell types that are likely to be relevant for a given trait;
- eCAVIAR and COLOC, two methods that quantify the probability that a variant is causal in both genetic association and eQTL studies;
- GREGOR, which integrates chromatin states with genetic associations derived from meta-analysis of the Knowledge Portal Network to generate p-values representing the significance of association between a tissue and trait;
- LD score regression (LDSR), which uses cell type-specific annotations and genetic association summary statistics in the Knowledge Portal Network to generate p-values representing the significance of association between a tissue and trait.
|Gene FOCUS table for PITX2|