R code to support "Genetic feature engineering enables characterisation of shared risk factors in immune-mediated diseases"

This contains R code used to generate figures and supplementary items reported in "Genetic feature engineering enables characterisation of shared risk factors in immune-mediated diseases"

Background Genome-wide association studies (GWAS) have identified pervasive sharing of genetic architectures across multiple immune-mediated diseases (IMD). By learning the genetic basis of IMD risk from common diseases, this sharing can be exploited to enable analysis of less frequent IMDs where, due to limited sample size, traditional GWAS techniques are challenging.

Methods Exploiting ideas from Bayesian genetic fine-mapping, we developed a disease-focused shrinkage approach to allow us to distil genetic risk components from  GWAS summary statistics for a set of related diseases. We applied this technique to 13 larger GWAS studies of common IMD, deriving a reduced-dimension `basis’ that summarised the multidimensional components of genetic risk. We used independent datasets including the UK Biobank to assess the performance of the basis and characterise individual axes. Finally we projected summary GWAS data for smaller IMD studies, with less than 1000 cases, to assess whether the approach was able to provide additional insights into genetic architecture of less common IMD or IMD subtypes, where cohort collection is challenging.

Results We identified 13 IMD genetic risk components. The projection of independent UK Biobank data demonstrated the IMD-specificity and accuracy of the basis even for traits with very limited case-size (e.g. vitiligo, 150 cases). Projection of additional IMD-relevant studies allowed us to add biological interpretation to specific components, e.g. related to raised eosinophil counts in blood and serum concentration of the chemokine  CXCL10 (IP-10). On application to 22 rare IMD and IMD subtypes we were able to not only highlight subtype-discriminating axes (e.g. for juvenile idiopathic arthritis) but also suggest eight novel genetic associations.

Conclusions Requiring only summary level data, our unsupervised approach allows the genetic architectures across any range of clinically-related traits to be characterised in fewer dimensions. This facilitates the analysis of studies with modest sample size by matching shared axes of both genetic and biological risk across a wider disease domain, and provides an evidence-base for possible therapeutic repurposing opportunities.