Background

In recent years there has been great progress developing genome-wide statistical tools for detecting causal variants, constructing prediction models and better understanding the genetic architecture of complex traits. However these tools use regression models involving very large numbers of predictors, and strong modelling assumptions are required to tackle the consequent problem of over-fitting. The results can be sensitive to these assumptions, and also to the effects of population structure, genotyping errors and the extent to which rare SNPs are included.

Course outline

I will cover mixed-model association analysis (e.g., Fast-LMM, GEMMA), risk prediction (e.g., polygenic risk scores, BLUP and MultiBLUP) and heritability analyses (GCTA, LDAK, LDSC), both using individual-level genetic data and summary statistics. I will emphasise the common elements of these methods, highlighting a standard framework that has emerged for genome-wide SNP analysis, while also contrasting the differences in underlying modelling assumptions.

The practicals will provide step-by-step details for analysing genetic data, starting either with individual-level data (e.g., PLINK files or the output from IMPUTE2) or summary statistics (p-values from a GWAS). There will be a selection of worked examples; to take part in the practicals, participants should bring a laptop with either MAC or LINUX OS

Prerequisites

Participants should have a basic understanding of statistics, and would ideally be familiar with the idea of a Bayesian regression model. In genetics, knowledge of SNP genotypes and linkage disequilibrium will be assumed. Computer scripts and output will be discussed that assume some familiarity with scientific computing using Linux. Experience with PLINK would be helpful but is not essential.

Provisional Timetable

09:15 - 11:45: Lecture 1 followed by Practical 1

Introduction to analysing GWAS data, including quality control, single-SNP analysis, polygenic risk scores, mixed-model analysis and gene-based analysis

11:45 - 12:45: Lunch (will be provided)

12:45 - 15:15  2 followed by Practical 2

Estimating heritability, confounding bias, bivariate correlations and enrichment of functional categories

Room details will be provided closer to the time; any questions, email doug@aias.au.dk

Registration

The course is free of charge but advance registration is required by sending an email with your name, department and institution/university to info@aias.au.dk by 13 March 2018.

Contact

Dr Doug Speed, AIAS Fellow, Aarhus Institute of Advanced Studies, Aarhus University

E-mail: doug@aias.au.dk

Høegh-Guldbergs Gade 6B

8000 Aarhus C

Denmark