Sparse Support Vector Machine for stratification of schizophrenia: Identifying novel biologically valid diagnostic categories to inform precision medicine

Employing Machine Learning algorithms is one way to explore the complex architecture of BIG genetic data and one such method, Support Vector Machines (SVM). The student will investigate the performance of sparse SVMs algorithm in simulated and real schizophrenia case/control data sets.

Current diagnostic categories do not map onto underlying biology and are at odds with the continuous nature of many psychiatric phenotypes. There is evidence for shared genetic risk across psychiatric disorders as well as genetic strata within disorders and our group have been at the vanguard of challenging existing categorical diagnostic classifications and of re-conceptualising the relationships between disorders [1]. However, the findings do not yet point to clinical strata that are useful for predicting outcomes or treatments, nor do they provide insights into the underlying biology of the strata.

The SZ group of the Psychiatric Genetics Consortium recently reported a meta-analysis [2] identifying 108 physically distinct genome-wide significant (GWS) loci. Tests for statistical interaction between all pairs of autosomal GWS index SNPs found no evidence for epistatic effects. It is possible that such effects could be present between variants outside GWS loci, or occur in the form of higher order interactions. Employing Machine Learning algorithms is one way to explore the complex architecture of BIG genetic data and one such method, Support Vector Machines (SVMs) [3]. This method can account for nonlinear genetic effects (interactions) via the use of Kernel functions in contrast to polygenic approaches which are implicitly additive, and for the large number of genetic variants via the use of sparsity parameters. The student will investigate the ability of sparse SVMs algorithm to discriminate between cases and controls using simulated data and real schizophrenia (SZ) case/control data. The best performing SVM models will be used to integrate environmental, clinical and molecular determinants of disease and develop etiological models to inform on novel approaches for prevention and therapy.

Aims of the project:
1. Investigate several approaches for SNP selection a) standard pruning and thresholding method, often employed in polygenic risk score (PRS) analysis; b) the mutual information (MI) approach which selects markers with largest MI with the outcome, etc.);
2. Compare the performance of the SVMs algorithm with different Kernels and sparsity parameters;
3. Compare the performance of the SVMs algorithm with logistic regression predictive models based upon polygenic risk scores;
4. Test the SVMs algorithm with different type of prediction variables (SNPs/genes/pathways);
5. Employ the best performing SVM models to integrate environmental, clinical and molecular data; 6. Provide pragmatic advice regarding the feasibility of prediction and stratification in SZ (environmental, clinical and genetic), and their optimal predictive accuracy.

Funding
This studentship is funded through GW4BioMed MRC Doctoral Training Partnership. It consists of full UK/EU tuition fees, as well as a Doctoral Stipend matching UK Research Council National Minimum (£14,553 p.a. for 2017/18, updated each year).

Additional research and training funding is available over the course of the programme. This will cover costs such as research consumables, courses, conferences and travel. Additional competitive funds are available for high-cost training/research.

The research project listed is in competition with 40 other studentship projects available across the GW4 BioMed MRC Doctoral Training Partnership. Up to 8 studentships will be awarded to the best applicants.

You will need to complete both an application to the GW4 BioMed MRC DTP for an ‘offer of funding’ and to Cardiff University for an ‘offer to study’.

Offer of Funding
Applicants will apply for funding via the centralised online application form, between 11th May and 9.30am 8th June 2017 (click link below).

Offer of Study
Applicants should submit an application for postgraduate study via the Cardiff University Online Application Service https://tinyurl.com/klqxt3s

Applicants should select Doctor of Philosophy (Medicine), with a start date of October 2017.

In the research proposal section of your application, please specify the project title and supervisors of this project and copy the project description in the text box provided. In the funding section, please select “I will be applying for a scholarship / grant” and specify that you are applying for advertised funding from GW4 BioMed MRC DTP.

If you are applying for more than one Cardiff University project, please note this in the research proposal section.