Integrative statistical analysis of -omics and GWAS data

class: top, right, title-slide

# Integrative statistical analysis of -omics and GWAS data
## Public defense
### Sina Rüeger
### 21 September 2018

---

---
class: inverse, left, middle

# Outline

- Introduction

- Results

- Discussion

---
class: inverse, left, middle

# Outline

- Introduction (= explaining the title)

- Results (= what I did during the PhD)

- Discussion (= did it work? & outlook)

---
class: inverse, center, middle

# Introduction

---
class: center, middle

## The goals to keep in mind: *Cornerstones of Medicine*

---
class: center, middle

## 1. What is "data analysis" and "statistics"?

---
class: left, middle

### Summarise data

#### Example: Distribution of human height in 100 individuals

![](slides_files/figure-html/unnamed-chunk-1-1.png)

---
class: left, middle

### Summarise data

#### Example: Distribution of human height in 100 individuals

![](slides_files/figure-html/unnamed-chunk-2-1.png)

- Mean: 164

- Standard deviation = 6.8

---
class: left, middle

### Identify relationships between A and B

#### Example: Relationship between age, sex and LDL (low-density lipoprotein)

.small[From: Freedman et al. (2004)]

---
class: left, middle

## Central principles of data analysis

- Data represents oftentimes an **approximation** for something we cannot measure *Use of BMI as a proxy for obesity*.

- The more data records we have, the **more precise** the estimation (of interest) *Relationship between sample size and the standard error*.

---
class: left, middle

## Keep in mind

- Statistical methods: **summarise data** & quantify relationships.

- Increasing **sample size** helps to estimate precisely.

---
class: center, middle

## 2. Bits of biology

or: what is *-omics*?

---
class: left, middle

## DNA 
.pull-left[
- is the blueprint of (human) life

- consists of nucleotides A, C, G and T 
]

.pull-right[
<img src="img/genes-DNA.png" width="800">
.small[Source: https://www.thoughtco.com/genetics-basics-373285]
]
---
class: left, middle

## Biological cascade

---
class: left, middle

## SNP (Single Nucleotide Polymorphism)

---
class: left, middle

## SNP (Single Nucleotide Polymorphism)

---
class: left, middle

.pull-left[
<blockquote class="twitter-tweet" data-lang="en">Two of my favourite things put together! <a href="https://t.co/1jE0hWr0Yt">pic.twitter.com/1jE0hWr0Yt</a>&mdash; Mike Inouye (@minouye271) <a href="https://twitter.com/minouye271/status/993073495205527552?ref_src=twsrc%5Etfw">May 6, 2018</a></blockquote>
<script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>
]

.pull-right[
## Linkage Disequilibrium (LD)

Parts of the **genome tend to be inherited together**, which creates sets of correlated SNPs in close proximity (‘in linkage disequilibrium (LD)’).

]

---
class: left, middle

## How DNA is measured

---
class: left, middle

## Keep in mind

- We are only interested in the varying part of the genome **SNPs**

- Parts of the genome are correlated **LD**

---
class: center, middle

## 3. GWAS (genome-wide association study)

---
class: left, middle

## Relationship between a genotype and trait/disease

![](slides_files/figure-html/unnamed-chunk-3-1.png)

---
class: left, middle

## Relationship between a genotype and trait/disease

![](slides_files/figure-html/unnamed-chunk-4-1.png)

---
class: left, middle

## Relationship between a genotype and trait/disease
<img src="img/gwas-a.png" width="600">

---
class: left, middle

## Relationship between MANY genotypes and trait/disease
<img src="img/gwas-b.png" width="600">

---
class: left, middle

## GWAS summary statistics

---
class: left, middle

## Large sample sizes needed

.pull-left[
<img src="img/maf_effectsize.png" width="400">
]

.pull-right[
- Solution \#1: Exploit genetic correlation and use chips (cheap) instead of sequencing (expensive)

- Solution \#2: Form a consortium
]

---
class: left, middle

## Example: GWAS in MDD

- [Wray et al. (2018)](https://doi.org/10.1038/s41588-018-0090-3)

- 44 novel loci associated with major depression disorder (MDD)

- a meta-analysis of GWAS summary statistics from 135'458 cases and 344'901 controls from six cohorts.

Resulting summary statistic was used:
- to identify the somatic tissues involved in MDD
- to estimate heritability explained by the new findings
- to identify biological pathways relevant to MDD
- to investigate the relationships with genetically correlated traits

---
class: left, middle

## Keep in mind

- Large sample sizes needed: **consortium**!

- Use of GWAS summary statistics with **follow-up methods**.

---
class: center, middle

## 4. Missing data

---
class: left, middle

## How to infer data that is missing?

---
class: left, middle

## How to infer data that is missing?

---
class: left, middle

## Basic principle

---
class: center, middle

## 5. Summary statistic imputation (SSIMP)

---
class: left, middle

## GWAS

---
class: left, middle
## Untyped variant

---
class: left, middle

## Genotype imputation

---
class: left, middle

## Genotype imputation

---
class: left, middle

## Summary statistic imputation

---
class: left, middle

## Main equation

---
class: inverse, center, middle

# Results

---
class: left, middle
# Output

1. Evaluation and application of summary statistic imputation 
 - *Evaluation and application of summary statistic imputation to discover new height- associated loci* (Rüeger, McDaid, and Kutalik 2018, PLOS Genetics)

1. Improving summary statistic imputation for mixed populations
 - *Improved imputation of summary statistics for mixed populations* (Rüeger, McDaid, and Kutalik 2017, BioRxiv)

1. Applications of summary statistic imputation
 - *Rare and low-frequency coding variants alter human adult height* (Marouli et al. 2017, Nature)
 - *Bayesian association scan reveals loci associated with human lifespan and linked biomarkers* (McDaid et al. 2017, Nature Communications)

1. Contributions to research in infectious diseases
 - *Impact of common risk factors of fibrosis progression in chronic hepatitis C* (Rüeger et al. 2015, Gut)
 - ...

1. Minor contributions to other publications

---
class: left, middle

---
class: left, middle

# Goals

1. **Compare** summary statistic imputation to genotype imputation.

1. **Test** the utility of summary statistic imputation on a real case on human height data.

---
class: left, middle

# 1. Compare SSIMP to genotype imputation

---
class: left, middle

---
class: left, middle

---
class: left, middle

# 2. Test SSIMP on height GWAS

---
class: left, middle

## Locuszoom plot of results

---
class: inverse, center, middle

# Discussion

---
class: top, middle

## Outlook for SSIMP

To impute summary statistics reliably the following is needed ☝️

### Large and diverse reference panels

- to improve of LD matrix estimation
- to impute low-frequency variants 
- to impute cosmopolitan/multi-ethnic GWASs

### Software & tooling

- easy to use and maintained

### Publicly available GWAS data

- the foundation of SSIMP

---
class: left, middle

## Conclusion

With my work on summary statistic imputation I have helped to **link incomplete GWAS** summary statistics with **follow-up methods** (such as multi-trait analyses).

<!---------
- **Improved SSIMP**: .small[improved imputation quality and optimised assembly of the LD matrix.]

- Quantified **performance & utility of SSIMP**: .small[compared to genotype imputation, identified groups of genetic variants that are hard to impute, and demonstrated in a case study the utility of summary statistic imputation.] 
----->

---
class: inverse, center, middle

.big[ Thank you! ]

Slides: [https://sinarueeger.github.io/publicdefense/slides#1](https://sinarueeger.github.io/publicdefense/slides#1)

Source code: [https://github.com/sinarueeger/publicdefense/](https://github.com/sinarueeger/publicdefense/)