HSM Polygenic score methodology in the emerging field of Polygenic Epidemiology

Over the last decade, researchers in the field of medical genetics have conducted a huge number of 'genome-wide association studies' (GWAS), which have identified thousands of genetic variants associated with hundreds of diseases, psychiatric disorders and human traits. While this endeavour has been exceptionally successful in highlighting regions of the human genome that contribute to human disease, it has also revealed that most human diseases are influenced by hundreds of genetic variants that each have a small impact on disease risk. This makes it extremely difficult to exploit an individual's genetic profile to predict how likely they are to contract different diseases or suffer adverse reactions to medical treatments, such as pharmaceutical drugs. As a result, genetics has yet to fulfil its greatest promise of initiating an era of stratified medicine: where disease preventions and treatments are individualised according to genetic profile.

However, methods are now being developed and applied to genetic data that take special account of the 'polygenic' nature of how genetics influences our disease risk. These methods are known collectively as 'polygenic score methods'. Based on their theoretical evaluation and practical application so far, early signs indicate that they may justify renewed hope in the potential of genetics to deliver stratified medicine. However, before this is possible, much theoretical work needs to go in to the development and testing of these polygenic score methods. A greater understanding of their performance is crucial, they need to be refined to produce more accurate disease risk prediction and extended to solve a wider variety of problems in medical genetics, and a set of strategies needs to be developed for how they can be exploited for stratified medicine. This proposal aims to achieve each of these goals.

This proposal begins with a detailed evaluation and comparison of the performance of different polygenic score methods. The results from this study, based on computer simulation, will offer insights into how these methods perform, what inferences they can make, and guide researchers on which method to use depending on their scientific question and study. The software tool that we will produce to perform this study will be made freely available as a web application, from which researchers will be able to: inspect results from our study presented in a customisable format, simulate and download their own polygenic data, and download and extend our simulation code to perform polygenic method evaluations and comparisons of their own.

Next the proposal introduces a set of new polygenic score methods, each one tailored to answer a specific scientific question. For example, while one method is designed to assess whether two diseases share a common genetic basis, useful for guiding which pharmaceutical drugs could be repurposed for other diseases, another helps determine whether an observed association between a risk factor and a disease is truly causal, and yet another infers the most likely direction of causality in such a relationship, which could answer key medical such as: Do high lipid levels increase the risk of Alzheimer's or does the onset of Alzheimer's result in higher lipid levels?

Finally, the proposal investigates a number of strategies for using polygenic score methods to aid in stratified medicine. We will outline and test approaches for using polygenic score methods to stratify heterogenous disorders into sub-sets of more biologically homogenous disorders, which will help make individual diagnosis, and thus subsequent treatment, more specific. We also describe ways in which polygenic scores on individuals can be used for more nuanced selection of participants in clinical trials, reducing their efficacy, cost, and ultimately leading to the development of drugs and treatments more tailored to individual genetic profile.

The latest analyses of GWAS data reveal a substantial polygenic component to most common human diseases. This poses a huge challenge to using genetics for phenotype prediction, and thus to effectuating stratified medicine. However, the development of prediction methods tailored to the polygenicity of human disease can transform results.

While the stringent GWAS significance threshold has ensured the discovery of genuine genotype-phenotype associations, it can be barrier to prediction. Phenotype prediction from genetic profile is typically optimised when applying a far more liberal threshold for SNP selection. While even polygenic prediction is inaccurate at an individual-level, its aggregation across samples enables a raft of applications that perform association testing.

The first polygenic score methods have now been proposed and applied to genome-wide SNP data. However, the increasing rate in their application is preceding careful examination of their relative power and accuracy. To address this, we perform the first comprehensive comparison study of their performance and build a simulation framework, accompanied by software tool and web application, to enable future benchmarking of polygenic methods across a consistent platform.

The polygenic risk score approach is the focus for method development here. Unlike the alternatives, PRS both exploits large-scale GWAS summary data and permits individual-level prediction. This combination will be key to realising stratified medicine as GWAS sample sizes grow and we outline a range of strategies to exploit the PRS approach for stratified medicine.

To date, only a single PRS method has been developed and utilised for a variety of applications. We introduce a suite of PRS methods, each tailored to their application. This sensitivity to the relevant scientific question is expected to yield substantial gains in power. Novel applications of PRS, such as testing direction of causality, are also proposed.

Drug development companies: Drug development is a huge national and international industry, which in recent years has recognised the lead that genes and genetic variants discovered in genome-wide association studies may provide for drug and therapeutic developments. A new field, pharmacogenetics, which primarily seeks to reduce the side-effects of developed drugs through knowledge of their genetic causes, has been borne out of the interest in genetic association studies. The focus of this project on disease risk prediction and on methods for using polygenic risk score methods for stratified medicine (Aim 3), which include approaches for disease stratification and on potential insights that could inform drug repurposing, should be of huge interest to drug development companies. Drug development companies are also likely to be hugely interested in any refinements of phenotype definition produces in our work into stratified medicine (Aim 3), and pharmaceuticals could gain immediate utility in particular from our research into using polygenic risk scores to perform screening in clinical trials.

Diagnostic companies: An important part of this project aims to produce new and refined phenotype definitions (Aim 3), which in many cases may necessitate or stimulate new measuring techniques, or an increased volume of previously used techniques, required for diagnosis. There has been a huge growth in molecular diagnostic companies since the emergence of GWAS in 2007, and more recently in the new area of 'companion diagnostics', where tests are performed that inform the efficacy of drugs either after or during drug development. Any additional insights into disease stratification and phenotype definition gained through the development of our polygenic risk score methods could be hugely beneficial to these companies.

Healthcare (NHS) policy makers: Similarly to diagnostic companies, any refinement of phenotype definition or diagnosis produced by this project could have a huge impact on policy decisions within the healthcare system, as well as on governmental public health advise based on epidemiological findings relating to new or refined phenotypes.

Clinicians: The stratified medicine aspect of this project in particular could be extremely interesting to clinicians working in the area of the related phenotypes and also by contributing to more accurate and systematic phenotype definition. More precise and systematic diagnoses can circumvent the problems of clinical subjective judgment in making clinical diagnostic decisions, which can in some cases lead to legislative procedures.