Grouping and tagging SNPs

Chris Wallace

�r Sys.Date()

Table of Contents

• 1. Grouping SNPs
• 2. The groups and tags classes
• 3. Automatic group building based on marginal posterior probabilities
• 4. Manual corrections

Chris Wallace // web // email

Grouping SNPs

We often (usually) cannot resolve an association to a single causal variant, and one of the aims of GUESSFM is to generate groups of SNPs, one of which is likely to be causal.

The groups and tags classes

The groups and tags classes are two ways to represent the same information, but use different structures which might be more convenient for one function or another.

library(GUESSFM)

## Warning: replacing previous import 'data.table::melt' by 'reshape::melt'
## when loading 'GUESSFM'

showClass("groups")

## Class "groups" [package "GUESSFM"]
## 
## Slots:
##                           
## Name:      .Data      tags
## Class:      list character
## 
## Extends: 
## Class "list", from data part
## Class "vector", by class "list", distance 2

showClass("tags")

## Class "tags" [package "GUESSFM"]
## 
## Slots:
##                           
## Name:      .Data      tags
## Class: character character
## 
## Extends: 
## Class "character", from data part
## Class "vector", by class "character", distance 2
## Class "data.frameRowLabels", by class "character", distance 2
## Class "SuperClassMethod", by class "character", distance 2
## Class "index", by class "character", distance 2
## Class "atomicVector", by class "character", distance 2
## Class "character or NULL", by class "character", distance 2
## Class "characterORconnection", by class "character", distance 2

Automatic group building based on marginal posterior probabilities

Once models have been fitted, we can average over them to find marginal posterior probabilities for each SNP. We make the assumption that SNPs in a group should display the following properties:

• the sum of MPI should be no more than 1
• SNPs in a group should not occur in the same model together very often
• SNPs in a group should be correlated (r²>0.5), and the MPI should tail off from the primary SNP with r²

snp.picker() aims to create such groups in an automated manner, given the summary of a GUESS run and a genotype dataset from which r² may be calculated. It returns an object of class snppicker which is an extension of the groups class and contains information to allow plotting of the relationship between r² and MPI that was used to define a group.

showClass("snppicker")

## Class "snppicker" [package "GUESSFM"]
## 
## Slots:
##                           
## Name:  plotsdata    groups
## Class:      list      list

Manual corrections

Plotting an object of class snppicker should be informative. The algorithm underlying snp.picker() has been designed so it is more likely to err by splitting a single group in two than by joining what might be two separate groups. If you suspect two groups should be merged, you can check the assumptions above using check.merge(), and perform a merge using group.merge().