Skip to contents

Filtering Genotype Tables

Brandon Monier

2023-11-08

Source: vignettes/genotype_filtration.Rmd
genotype_filtration.Rmd

Introduction

Genotype table overview

In this vignette, we will discuss how you can filter genotype tables in rTASSEL. As a reminder, a genotype table in TASSEL is essentially an \(n \times p\) matrix consisting of \(n\) samples or taxa with \(p\) genotypic observations. To illustrate this, see the diagram below:

In this illustration, sequence information is processed for variant sites of each parental copy. Variant data is partitioned into an \(n \times p\) matrix for each sample and relevant variant marker information (e.g. 0:2). This includes:

  • The TASSEL index for each variant site (<this value> :)
    • Note: since TASSEL is coded in Java, this value will start at 0!
  • The physical position (bp) for each variant site (: <this value>)

Genotype information (e.g. each \(n \times p\) element) can be viewed in several ways. The default view (as shown above) reflects the following:

Cell Type Definition Color
Major alleles The most frequently observed allele of each column Yellow
Minor alleles The least frequently observed allele of each column Blue
Heterozygous Sites where parental copies differ (e.g. C:G) Green
Unknown Sites where no information is given (e.g. NAs) Grey/White

Filtering basics

Since TASSEL allows for many features to be represented in a genotype table, rTASSEL’s filtration parameters are numerous. In the following sections, we will discuss how each set of parameters work. In general, these filtration schemes can be applied for samples (i.e. rows) and marker information (i.e. columns).

The two main methods to accomplish this are:

For each section example code will be given with an illustrated genotype table to explain each method. Prior knowledge about rTASSEL data structures is needed so please refere to the vignette Getting Started with rTASSEL if you are new to this package!

Note: We will be using the the magrittr pipe operator %>% for the following examples. More information about this can be found here.

Filtering by sites…

Overview

In order to filter by variant sites (e.g. marker information), we can use the function filterGenotypeTableSites(). This function was modeled after TASSEL’s FilterSiteBuilder plugin to ensure familiarity between the two methods. The parameters can be broken up into several components:

  • Genotype information
    • siteMinCount
    • siteMinAlleleFreq
    • siteMaxAlleleFreq
    • minHeterozygous
    • maxHeterozygous
  • Indexed variant sites
    • startSite
    • endSite
  • Physical marker positions
    • startChr
    • endChr
    • startPos
    • endPos
  • R objects
    • gRangesObj
  • External files
    • bedFile
    • chrPosFile

Note: While the following examples will illustrate these parameters in simplified formats, genotype information parameters and marker index and position parameters (i.e. indexed variant sites, physical marker positions, R objects, or external files) can be passed simultaneously!

Let’s assume we have an rTASSEL data object that contains a genotype table:

myGT
## A TasselGenotypePhenotype Dataset
##   Class.............. TasselGenotypePhenotype 
##   Taxa............... 5 
##   Positions.......... 5
##   Taxa x Positions... 25
## ---
##   Genotype Table..... [x]
##   Phenotype Table.... [ ]

Let’s also assume it contains the same structure as the prior illustration:

In order for us to understand how the next few parameters work, we will illustrate how allelic frequencies are calculated:

…using allelic frequency

If we want to filter by minor allele frequency, we can use the following parameters…

myGT %>% 
    filterGenotypeTableSites(
        siteMinAlleleFreq = 0.3,
        siteMaxAlleleFreq = 1.0
    )

…which will remove the following columns:

We can also set a range using a limit with the siteMaxAlleleFreq parameter…

myGT %>% 
    filterGenotypeTableSites(
        siteMinAlleleFreq = 0.2,
        siteMaxAlleleFreq = 0.3
    )

…which will remove the following columns:

…using NA data

We can also filter by missing sites (encoded in TASSEL with N). Similar to how we calculated frequency of minor alleles, we can apply the same logic with missing data with the siteMinCount parameter:

myGT %>% 
    filterGenotypeTableSites(
        siteMinCount = 5
    )

This parameter can range from 0 to the number of rows (e.g. taxa) within your genotype table. Since our example data only has 5 rows, this prior example will filter out any markers that contain at least 1 missing variant marker site:

…using heterozygosity frequency

Similar to the prior two sections, we can filter by calculated frequency of heterozygous sites using the parameters, minHeterozygous and maxHeterozygous. For example, if we wanted to filter sites containing \(>\) 10% heterozygous sites, we could use the following:

myGT %>% 
    filterGenotypeTableSites(
        minHeterozygous = 0.0,
        maxHeterozygous = 0.1
    )

which would filter out sites with calculated frequencies:

…using variant indexes

For the next several sections we will be illustrating how to filter sites using positional information for each marker. To better explain these concepts, let’s modify our genotype table (myGT) to showcase (1) a broader range of positions and (2) different sequence IDs (i.e. chromosomes). Note Chromosome IDs are not illustrated this way in TASSEL and are only for demonstration purposes.

To filter by site index, we must first set the siteRangeFilterType parameter to sites. Next, we will be able to use the parameters, startSite and endSite:

myGT %>% 
    filterGenotypeTableSites(
        siteRangeFilterType = "sites",
        startSite = 1,
        endSite = 3
    )

In our example above, this will retain indexed sites 1 through 3:

…using physical positions

To filter by physical positions, we must first set the siteRangeFilterType parameter to position. Next we will be able to use the parameters:

  • startChr
  • startPos
  • endChr
  • endPos

The *Chr parameters will specify which chromosomes or sequence IDs will remain and *Pos parameters will specify the physical positions within each chromosome sequence. For example, if we wanted to filter just chromosome 1, we can specify the following parameters:

myGT %>% 
    filterGenotypeTableSites(
        siteRangeFilterType = "position",
        startChr = 1,
        endChr = 1,
        startPos = NULL,
        endPos = NULL
    )

If we want to be more specific in our filtering, we can include positional information as well:

myGT %>% 
    filterGenotypeTableSites(
        siteRangeFilterType = "position",
        startChr = 1,
        endChr = 2,
        startPos = 250,
        endPos = 700
    )

…using GRanges objects

In the prior example, we have manually set parameters to filter by physical position. The caveat to this process is that you can only specify a “fixed” range of start and stop positions. For more dynamic filtering options, we can specify either external files (next section) or range-based R objects.

GRanges class objects from the GenomicRanges package serve as the standard data structure for representing genomic locations within the Bioconductor project. A basic representation of this class entails the following data types:

  • seqnames - an Rle object of sequence IDs (e.g. chromsomes)
  • ranges - an IRanges object of start and stop coordinates for range info
  • strand - an Rle object representing sequence strand info (+, -, or *)

For our example, we will represent a GRanges object as the variable gr:

gr
## GRanges object with 3 ranges and 0 metadata columns:
##       seqnames    ranges strand
##          <Rle> <IRanges>  <Rle>
##   [1]     chr1   250-500      *
##   [2]     chr2   213-400      *
##   [3]     chr2   500-700      *
##   -------
##   seqinfo: 2 sequences from an unspecified genome; no seqlengths

To implement this in our filtration method, we can pass this GRanges object into our gRangesObj parameters and setting siteRangeRilterType to "none":

myGT %>% 
    filterGenotypeTableSites(
        siteRangeFilterType = "none",
        gRangesObj = gr
    )

Which will dynamically filter the following marker indexes:

…using BED files

rTASSEL allows filtering for both external BED files and chromosomal position files. Similar to our GRanges object, BED files can also represent genomic range information but as a flat file. In our example, let’s take the same range information and save it as a BED file:

$ cat my_ranges.bed

chr1    250   500
chr2    213   400
chr2    500   700

This external file can be passed to our filtration method by setting siteRangeFilterType to "none" and entering the location of the file on disk in the bedFile parameter:

myGT %>% 
    filterGenotypeTableSites(
        siteRangeFilterType = "none",
        bedFile = "my_ranges.bed"
    )

This will filter our genotype table as previously shown in GRanges section:

…using chromsomal position files

Similar to a BED file, we can also pass a chromosomal position file. This is a tab-separated file of two columns:

  • Chromosome - the sequence or chromosome ID for each marker
  • Position - the physical position (bp) of each marker

In our example, this would look something like this:

$ cat my_chr_pos.tsv

Chromsome   Position
1   300
2   213
2   665

This external file can be passed to our filtration method by setting siteRangeFilterType to "none" and entering the location of the file on disk in the chrPosFile parameter:

myGT %>% 
    filterGenotypeTableSites(
        siteRangeFilterType = "none",
        bedFile = "my_chr_pos.tsv"
    )

This will also filter our genotype table as previously shown in GRanges section:

Filtering by taxa

Overview

In the prior sections we have discussed how to filter \(p\) observations (i.e.  columns). In the following sections, we will illustrate ways to filter \(n\) rows. In order to filter by variant sites (e.g. marker information), we can use the function filterGenotypeTableTaxa(). This function was modeled after TASSEL’s FilterTaxaBuilder plugin to ensure familiarity between the two methods.

While not as complex as site filtration, taxa filtration parameters can be seperated into the following sections:

  • Genotype information
    • minNotMissing
    • minHeterozygous
    • maxHeterozygous
  • R objects
    • taxa

Since we will be filtering by taxa, let’s represent our toy illustration to contain actual maize taxa IDs instead of colorful circles. This is similar to how this would be encoded in TASSEL:

…using NA data

Similar to site filtration, we can filter by missing sites (encoded in TASSEL with N). Instead of column (e.g. marker) filtration, this will filter taxa (rows) based on missing data. Similar to how we calculated frequency of minor alleles, we can apply the same logic with missing data with the minNotMissing parameter:

myGT %>% 
    filterGenotypeTableTaxa(
        minNotMissing = 1.0
    )

This parameter is based on the “missingness” frequency calculated for each row in your genotype table. Since our example data only has 5 rows, this prior example will filter out any taxa that contain at least 1 missing variant marker site:

…using heterozygosity frequency

We can also filter taxa by calculated frequency of heterozygous sites using the parameters, minHeterozygous and maxHeterozygous. For example, if we wanted to filter all taxa that contain at least one heterozygous site, we could employ the following parameters:

myGT %>% 
    filterGenotypeTableTaxa(
        minHeterozygous = 0.0,
        maxHeterozygous = 0.0
    )

which would filter out taxa with calculated frequencies:

…using R objects

Filtering select taxa IDs can be achieved by passing an atomic vector of character data types to the taxa parameter. In our example, we will select the following IDs:

myFavTaxa
## [1] "B73"  "B97"  "Ky21"

This vector can passed to the following argument:

myGT %>% 
    filterGenotypeTableTaxa(
        taxa = myFavTaxa
    )

Which will filter out any IDs that are not contained in the previous vector object:

This vector logic can applied to various types of subsetting methods in R. For example, if we had all of the taxa represented in one vector:

allTaxa
## [1] "B73"   "B97"   "HP301" "Ky21"  "Mo17"

We could filter from this object using regular expressions. One common method is using the str_subset method from the stringr package in conjunction with the magrittr pipe operator %>%:

myGT %>% 
    filterGenotypeTableTaxa(
        taxa = allTaxa %>% str_subset("^B|^Ky")
    )

This will create a subsetted vector containing our taxa IDs of interest in the prior example:

While this example might be rather basic, this method is quite powerful if dealing with a massive breeding panel of several hundred samples.

Piping methods together

Example 1

Like with all good practices in R and programming, various rTASSEL methods can be “piped” together in one continuous flow of steps. For example, we could filter simultaneously on sites and taxa using both methods in conjunction with the magrittr pipe operator %>%. Let’s use our previous toy data set:

Let’s (1) select taxa IDs that start with the letter “B” or “K” and (2) select indexed TASSEL marker sites 1 through 3. This will filter elements shown below:

myGT %>%
    filterGenotypeTableTaxa(
        taxa = allTaxa %>% str_subset("^B|^Ky")
    ) %>% 
    filterGenotypeTableSites(
        siteRangeFilterType = "sites",
        startSite = 1,
        endSite = 3
    )

Example 2

We can also implement the same methods with other rTASSEL methods. In this example we will (1) select taxa IDs that start with the letter “B” or “K”, (2) select indexed TASSEL marker sites 1 through 3, and (3) export to a VCF flat file:

myGT %>%
    filterGenotypeTableTaxa(
        taxa = allTaxa %>% str_subset("^B|^Ky")
    ) %>% 
    filterGenotypeTableSites(
        siteRangeFilterType = "sites",
        startSite = 1,
        endSite = 3
    ) %>% 
    exportGenotypeTable(
        file = "my_filtered_gt.vcf"
        format = "vcf"
    )