Abstract P5-12-16: Combinatorial risk scores: Personalized multi-omic prediction of disease risk

Abstract

Abstract Complex traits with heritable and lifestyle components present a challenge for traditional polygenic risk scores. These aggregate disease risk associations from multiple (tens to millions) low effect size SNVs to provide a quantitized group level risk. Typically this results in a binomial distribution of risk with the top and bottom percentiles having relatively clear clinical guidelines and a large number of intermediate patients for whom guidance is unclear. From previous work on the genetic architecture of breast cancer (http://bit.ly/2JHLk8q) it is clear that particular combinations of multiple SNPs (epistatic interaction signatures) are significantly associated with specific levels of disease risk and disease protective effects. We have now extended identification of these signatures to include both genomic and phenotypic factors. We found: · 3,045 novel risk signatures (combinations of 5-13 SNP variants) that occurred in large numbers of patients and zero controls · 5 novel protective disease signatures (combinations of 2-10 SNPs associated with reduced disease risk) · 10 genetically non-overlapping patient cohorts, each with a different disease risk · Disease risk signatures that combine genetic and phenotypic factors, such as ethnicity, obesity, drinking and smoking We have worked with UK Biobank, CIMBA and BCAC datasets to evaluate the risks/protective effects associated with each of these signatures. From these we are building combinatorial risk scores that better predict total lifetime disease risk, age of onset and therapy response. In the context of an individual patient, their specific SNVs and phenotype data can now be used to construct a personalized combinatorial risk score, which has the potential to enable both better diagnosis and theranostics. New results from these efforts will be presented at SABCS. GWAS aim to find (single) genetic variant loci associated with specific phenotypes. While GWAS has been useful at identifying disease associated factors, it is known to provide only a limited model for explaining complex diseases as very few loci have significant effect sizes and most diseases are highly polygenic (Boyle, 2017). In addition, GWAS cannot directly include the impact of non-genomic factors such as phenotype, lifestyle and comorbidities that may modulate disease processes and exert significant influence over disease risks. Current detection methods for disease associated combinations of SNPs (epistatic interactions) are able only to find combinations of two or at most three SNPs from a preselected list. Here, we present an alternative to the single-locus limitations of GWAS. Penetrance for Clusters (sets of non-redundant states) validated using different FDRs for BRCA2 dataset(FDR) False Discovery Rate %# of Clusters# Layers# of SNP Genotypes# of CasesPenetrance %204045-132,11379950.7102355-131,32062739.851575, 7-1386851332.61327, 10-1314222114.0Analysis using a False Discovery Rate (FDR) of 5% identified 3,045 states (unique n-SNP combinations) at layers (order) 5 and 7–13 that were found to differentiate breast cancer susceptibility. The penetrance in the cohort depends on the FDR used as shown in Table. Citation Format: Uckun S. Combinatorial risk scores: Personalized multi-omic prediction of disease risk [abstract]. In: Proceedings of the 2018 San Antonio Breast Cancer Symposium; 2018 Dec 4-8; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2019;79(4 Suppl):Abstract nr P5-12-16.