The use of multiple SNP data to predict schizophrenia risk

Jules Hernandez-Sanchez, Cameron Hurst, Dimitrios Vagenas, Christopher D Swagell, Ian Hughes, Bruce R Lawford, Ross Young, Charles Phillip Morris, Joanne Voisey

Research output: Contribution to specialist publicationArticle


Schizophrenia affects one percent of the population and has life-long debilitating consequences for those affected. It is a complex genetic disorder that results from the interaction of multiple gene variants and environmental factors. To date, numerous polymorphisms have been identified that are associated with schizophrenia but it is clear that no single polymorphism can accurately predict schizophrenia status.

Using a multiple candidate gene approach, 30 candidate genes were selected to test association with schizophrenia. Single nucleotide polymorphisms
(SNPs) in these genes were selected because they were either Hap Map tag-SNPs or because of their location in a functional gene domain. Initially, a total
of 273 SNPs were genotyped in 160 DSM-IV diagnosed schizophrenia patients and 250 control samples. After quality control, 151 SNPs in 29 different
genes were used in this study. In order to evaluate the best method to combine multiple SNP data, five different statistical classifiers were used to predict
schizophrenia risk. The five classifiers evaluated were; binary logistic regression (BLR), support vector machines (SVM), decision trees (DT), adaptive boosting
(AB), and partial-least-squares with linear-discriminant-analysis (PLS-LDA).

The best classifier was BLR but it was more informative to use several classifiers. The synonymous SNP, rs7301328, in the glutamate receptor 2B gene
(GRIN2B) was consistently selected among several classifiers (BLR, DT and AB). All classifiers utilised main effects of SNPs only but given that all genes were
functional candidates for schizophrenia, we hypothesised they may interact. As BLR was the best classifier, we used it to estimate direct and interaction
effects between all pair-wise combinations of SNPs. Additive-additive, additive-dominant and dominant-dominant interactions that averaged z-scores >
3.5 are reported. The greatest number of additive-additive interactions involved the catechol methyl Transferase (COMT) gene but both COMT and dopa
decarboxylase (DDC) showed a large number of dominant-dominant interactions and DDC was over-represented in terms of additive-dominant interactions.
While this panel of SNPs does not have the required sensitivity or specificity to be used as a diagnostic test, it is anticipated that the approach described in
this study will lead to a test for improved early diagnosis of schizophrenia. Such a test will enable early intervention strategies with the ultimate objective of
preventing schizophrenia onset and progression.
Original languageEnglish
Number of pages7
Specialist publicationJSM Schizophrenia
Publication statusPublished - Mar 2017
Externally publishedYes


Dive into the research topics of 'The use of multiple SNP data to predict schizophrenia risk'. Together they form a unique fingerprint.

Cite this