M20 - COMBINING POLYGENIC RISK SCORES ACROSS SEVERAL TRAITS CAN IMPROVE SCHIZOPHRENIA RISK PREDICTION.

Schizophrenia is a highly heritable disorder and is known to have a substantial common polygenetic component [1]. The polygenic risk score approach [1] allows common genetic liability to the disorder to be directly estimated in individuals regardless of their affected status. In addition to weakly predicting schizophrenia affected status in independent case-control datasets, polygenic risk score approaches have revealed significant shared SNP heritability between schizophrenia and multiple psychiatric and behavioural traits [2 , 3]. We therefore postulated that combining risk score profiles derived from several traits might improve the prediction of schizophrenia risk. Polygenic risk score profiles were calculated for the CLOZUK schizophrenia GWAS study [4] (11260 cases, 24542 controls) by training on six recent GWAS datasets: schizophrenia (independent of the CLOZUK set), bipolar disorder, bipolar vs schizophrenia [5] , major depression disorder, educational attainment [6] and neuroticism [7]. Principal component analysis (PCA) was employed to generate factor loadings for the six polygenic risk score profiles. The advantage of PCA is that it converts a set of possibly correlated variables (e.g. due to shared genetic liability) into a set of uncorrelated variables, with the first principal component accounting for the greatest degree of variance in the dataset outcome. We investigated the prediction accuracy of the first principal component compared to the schizophrenia polygenic risk score by means of area under the curve (AUC) analysis. Our analysis indicates that schizophrenia case prediction accuracy of the combined schizophrenia and neuropsychiatric-related first principal component is increased as compared to the AUC of schizophrenia polygenic risk scores alone. The first principal component also visually separates cases and controls in the CLOZUK individuals based upon their combined (schizophrenia and neuropsychiatric-related traits) profiles. We have proposed an approach to combine genetic evidence for schizophrenia and neuropsychiatric-related traits into one score and tested its prediction accuracy. This approach takes advantage of the shared genetic risk between schizophrenia, psychiatric disorders and behavioural traits, and at present, can improve the prediction of schizophrenia risk. [ABSTRACT FROM AUTHOR]