Filters for Next Generation Sequencing

# FiNGS: Filters for Next Generation Sequencing

## Key features

• Filters SNVs from any variant caller to remove false positives
• Calculates metrics based on BAM files and provides filtering not possible with other tools
• Fully user-configurable filtering (including which filters to use and their thresholds)
• Option to use filters identical to ICGC recommendations

## Introduction

Somatic variant callers compare matched pairs of tumor-normal samples to produce variant calls. The results can be extremely rich in false positives due to confounding factors such as the purity of the samples, artifacts introduced by sequencing chemistry, the alignment algorithm and the incomplete and repetitive nature of reference genomes.
It has become common practice to attempt to ameliorate these effects using a variety of filtering techniques, including taking the intersect of results from multiple variant callers and employing some post-calling filtering. This ad-hoc filtering varies greatly between laboratories. Attempts have been made to standardize the methodology of this filtering, with recommendations produced by the International Cancer Genome Consortium (ICGC) (Alioto et al., 2015).
We have developed Filters for Next Generation Sequencing (FiNGS), software written specifically to address these filtering issues. FiNGS can implement the ICGC filtering standards and has filters and thresholds that are fully configurable, which substantially increases the precision of results and provides high quality variants for further analysis.

## Availability

FiNGS is open source and released under the Apache License, Version 2.0. The latest source code is freely available at GitHub.

Python 3

## Quickstart guide, with example data and test

You have three options for running FiNGS;

### Bioconda installation

Create new conda environment Install package (and dependencies) Run example

### Standalone installation via GitHub

FiNGS depends on Python3 and several packages. We recommend using a Conda environment to manage Python and associated packages. On Debian-based systems such as Ubuntu, you can install Python3 and pip, the recommended tool for installing Python packages using the following commands:

apt-get -y install python3
apt-get -y install python3-pip


Now Python3 and pip are installed, you may install the required Python packages:

pip3 install PyVCF
pip3 install pysam
pip3 install editdistance
pip3 install scipy
pip3 install joblib


Clone the git repository and run the included example script in the exampledata directory.

git clone https://github.com/cpwardell/FiNGS.git
cd FiNGS/exampledata
./test.sh


You should see the progress printed to the screen. When filtering is complete, a results directory will have been created containing the output.

### Docker installation

This guide assumes you have Docker installed and have some basic knowledge on how to use it. The Dockerfile builds an image based on Ubuntu 16.04. You need to get a copy of the Dockerfile in this repository; below we use "wget" on Linux to download it, but you could just as easily copy and paste the link in your web browser and "right click/save as" the file. The Docker build command works identically in both Bash on Linux and PowerShell on Windows and assumes that you're in the same directory as the dockerfile named "Dockerfile".

# Download the Dockerfile from this address:
wget https://raw.githubusercontent.com/cpwardell/FiNGS/master/Dockerfile
# Build the image and call it "fings" (lowercase)
docker build -t fings .


You can test that your image works by running a container interactively:

docker run -it fings
cd /FiNGS/exampledata
./test.sh


When you run it on your own data, you can mount the location of your files as below. This would output a results directory in the directory the command was executed in. Note that the -u argument ensures that files created by the docker container will be owned by the user invoking the docker.

docker run -v /path/to/tumorbamdir:/tumorbamdir -v /path/to/normalbamdir:/normalbamdir -v /path/to/vcfdir:/vcfdir -v $PWD:/local -w /local -u$UID:\$UID fings /bin/bash -c "python3 /FiNGS/FiNGS.py -n /normalbamdir/normal.bam -t /tumorbamdir/tumor.bam -v /vcfdir/somatic.vcf --PASSonlyin --PASSonlyout"


## Suggested usage

• Use default filtering thresholds
• Use every available processor
• Only consider variants with a PASS filter value from the variant caller
• Only emit variants that PASS all filters
python3 /path/to/FiNGS.py -n /path/to/normal.bam -t /path/to/tumor.bam -v /path/to/somaticvariants.vcf --PASSonlyin --PASSonlyout

• ICGC mode:
python3 /path/to/FiNGS.py -n /path/to/normal.bam -t /path/to/tumor.bam -v /path/to/somaticvariants.vcf -r /path/to/reference/genome.fasta --PASSonlyin --PASSonlyout --ICGC


## Basic usage and outputs

In the simplest case, provide FiNGS with paths to a VCF from a somatic variant caller and the BAM files used to produce it.

python3 FiNGS.py -n /path/to/normal.bam -t /path/to/tumor.bam -v /path/to/somaticvariants.vcf


FiNGS will create a directory called "results" containing the following files:

• inputvcf.filtered.vcf
A VCF containing the filtered results. Descriptions of the filters used and their threshold values are stored in the header, and PASS/fail status stored in the FILTER column of each record
• plots.pdf
Plots for every filter applied in a single PDF. The first page shows a table of the PASS/Fail counts for each filter. Subsequent pages show kernel density plots of the data used, with a dashed vertical line demarcating the threshold used; the red region failed the filter, the green region passed
• log.txt
The log file for the run. Contains the command line arguments used and a complete log of the run
• filterresults.txt.gz
A gzipped text file giving the results of each filter for every variant
• tumor.combined.txt.gz and normal.combined.txt.gz
All metrics collected and used for filtering are stored in these gzipped text files. A dictionary explaining the contents of each column is below
• summarystats.txt.gz A gzipped text file containing summary stats that may be used for filtering

• -v Required; path to the somatic variant VCF from any variant caller
• -t Required; path to the tumor BAM file used to produce the VCF
• -n Required; path to the normal BAM file used to produce the VCF
• -r Optional; absolute path to faidx indexed reference genome; required if using 'repeats' filter
• -d Optional; path to output directory. Default is to create a directory called "results" in the current working directory
• -p Optional; path to file specifying filtering parameters. Details on filters and default values is provided below. Default is a file located at FiNGS/filter_parameters.txt
• -c Optional; chunk size, the number of variants to process per chunk. Default is 100
• -m Optional; maximum read depth. Reads beyond this depth will be ignored and a warning emitted to the log file. Default is 1000
• -j Optional; number of processors to use. -1 uses all available. Default is -1
• --ICGC Optional; use filters identical to those recommended by the ICGC (Alioto et al, 2015). File is located at FiNGS/icgc_filter_parameters.txt
• --logging Optional; change logging level. Default is INFO, can be DEBUG for more detail or NOTSET for silent
• --overwrite Optional; allow overwriting of existing results if rerunning
• --PASSonlyin Optional; only consider variants with a PASS in the filter field of the input VCF
• --PASSonlyout Optional; only write variants that PASS all filters to the output VCF

## Getting help

Run FiNGS with no additional arguments to get the help file. If there's something not adddressed here, or if you need further help, raise an issue on GitHub or find me online.

## Citing FiNGS

A paper is being prepared for submission shortly and will be referenced here when available.

## Description of filters

FiNGS assesses variants using any combination of these possible filters. Below is a table describing them, their default thresholds and ICGC thresholds. NA values mean that the filter is not employed in eithe the default or ICGC mode. Users can create their own parameter file using any combination of filters and thresholds.

Filter name Description Default value ICGC value
minaltcount Minimum number of ALT reads in tumor 3 4
minbasequality Minimum median base quality (separate filters for ALT reads in tumor, REF reads in tumor and REF reads in normal) 30 30
minmapquality Minimum median mapping quality of ALT reads in tumor 50 40
minmapqualitydifference Maximum difference between median mapping quality of ALT reads in tumor and REF reads in normal 5 5
enddistance Maximum median shortest distance to either aligned end in tumor 10 10
zeroproportion Maximum proportion of zero mapping quality reads in tumor and normal 0.05 0.1
minimumdepth Minimum depth in tumor and normal 10 NA
maximumdepth Maximum depth in tumor and normal 1000 NA
minvaftumor Minimum VAF in tumor 0.05 NA
maxvafnormal Maximum VAF in normal 0.03 NA
maxoaftumor Maximum OAF in tumor 0.04 NA
foxog FoxoG artifact proportion (see note below) 0.9 NA
editdistance Maximum edit distance of ALT reads in tumor (maximum edit distance of REF reads in tumor is 1 less than this value) 4 NA
maxsecondtumor Maximum proportion of secondary alignments in tumor 0.05 NA
strandbiasprop Strand bias exclusion proportion (see note below) 0.1 NA
strandbiassimple Maximum strand bias (see note below) NA 0.02
maxaltcount Maximum number of ALT reads in normal NA 1
snvcluster50 Maximum number of mutations within 50 bp (see note below) NA 2
snvcluster100 Maximum number of mutations within 100 bp (see note below) NA 4
repeats Maximum length of 1/2/3/4mer repeats around the variant position (see note below) NA 12
• Note on foxog filter
C>A|G>T variants can be oxidation artifacts introduced during library preparation (Costello et al, 2013). These "OxoG" artifacts have a telltale read orientation, with the majority of ALT reads in the artifact orientation. All C>A|G>T variants are classified as being part of two binomial distributions, one centered at 0.5 (50% artifact orientation, 50% non-artifact orientation) and the other at the filter value (defautl is 0.9, which is 90% artifact orientation reads). C>A|G>T variants classified as OxoG are removed.

• Note on strandbiasprop filter
Strand bias is defined below (Guo et al., 2012) where a,c represent the forward and reverse strand allele counts of REF reads and b,d represent the forward and reverse strand allele counts of ALT reads. The topmost proportion of biased variants is removed (e.g. suggested value is 0.1, leading to the top 10% variants with the highest strand bias being removed. Note that this is not the GATK strand bias, which is calculated differently. Note that this filter is available, but is not implemented in either of the default settings.
Strand bias: |b/(a+b)-d/(c+d)|/ ((b+d)/(a+b+c+d))
GATK strand bias: max(((b/(a+b))*(c/(c+d)))/((a+c)/(a+b+c+d)),((d/(c+d))*(a/(a+b)))/((a+c)/(a+b+c+d)))

• Note on strandbiassimple filter
Maximum strand bias. This is defined very simply as the minimum proportion of reads in either direction. e.g. if there were 100 reads and only 1 were forward, strand bias would be 0.01. Strand bias: min(forward/(forward+reverse),reverse/(reverse+forward)) or min((a+b)/(a+b+c+d),(c+d)/(a+b+c+d))

• Note on SNVcluster50 and SNVcluster100 filters
Maximum number of SNVs in the input VCF in a 50 or 100 bp window centered on the current SNV. SNVs must have at least 2 reads supporting them and a VAF>=5%.

• Note on repeats filter
Maximum length of 1/2/3/4mers surrounding the SNV. Lengths must be factors of repeats e.g. n=8 would only consider 1/2/4mers because 3 is not a factor of 8. Repeated regions frequently result in false positive variants. When using this filter, a reference genome must be supplied so FiNGS can find the flanking sequences.

## Dictionary of values reported in the metrics files

The gzipped tumor.combined.txt.gz and normal.combined.txt.gz output files contain all metrics calculated for every input variant. Each row is a single variant. Non-integer values are rounded to 3 decimal places. Not all of the values reported are used for filtering.

Column Description Example
UID Unique Identifier for variant 1:931362:G:A
CHR Chromosome 1
POS Position on chromsome 931362
REF Reference allele G
ALT Alternate allele A
refcount Count of REF alleles 100
altcount Count of ALT alleles 19
varianttype SNV or INDEL SNV
depth Depth of all reads 122
vaf Variant Allele Frequency (altcount/depth) 0.156
raf Reference Allele Frequency (refcount/depth) 0.82
oaf Other Allele Frequency ((depth-altcount-refcount)/depth) 0.025
medianbaseq Median base quality (all reads) 32
medianbaseqref Median base quality (REF reads only) 34.5
medianbaseqalt Median base quality (ALT reads only) 32
medianmapq Median mapping quality (all reads) 60
medianmapqref Median mapping quality (REF reads only) 60
medianmapqalt Median mapping quality (ALT reads only) 60
zeros Total number of zero mapping quality reads 0
zerospersite Proportion of reads that have zero mapping quality 0
goodoffsetproportion Proportion of variants that occur within the first 2/3rds of the mean read length 0.664
mad1 Median absolute deviation of distancetoend1median 34
mad2 Median absolute deviation of distancetoend2median 34.5
distancetoend1medianref Median distance to lefthand soft-clipped read end (REF reads only) 76
madref1 Median absolute deviation of distancetoend1medianref 31.5
distancetoend2medianref Median distance to righthand soft-clipped read end (REF reads only) 64
madref2 Median absolute deviation of distancetoend2medianref 29
distancetoend1medianalt Median distance to lefthand soft-clipped read end (ALT reads only) 65
madalt1 Median absolute deviation of distancetoend1medianalt 25
distancetoend2medianalt Median distance to righthand soft-clipped read end (ALT reads only) 85
madalt2 Median absolute deviation of distancetoend2medianalt 25
shortestdistancetoendmedian Median of shortest distance of distancetoend1alt and distancetoend2alt 42
madaltshort Median absolute deviation of shortestdistancetoendmedian 19
sb Strand bias, see definition above 0.22
gsb GATK strand bias, see definition above 0.185
fishp P value for Fisher's exact test for strand bias 0.614
FR Count of forward reads supporting REF allele 36
FA Count of forward reads supporting ALT allele 8
RR Count of reverse reads supporting REF allele 64
RA Count of reverse reads supporting ALT allele 11
altsb Simple strand bias of ALT reads 0.421
refsb Simple strand bias of REF reads 0.36
allsb Simple strand bias of all reads 0.37
F1R2 Count of reads in FoxoG orientation 1 8
F2R1 Count of reads in FoxoG orientation 2 11
FoxoG Oxoguanine artifact orientation proportion, only relevant for for C>A or G>T mutations, defined in Costello et al, 2013 0.421
refld Edit distance for REF reads 2
altld Edit distance for ALT reads 3
refsecondprop Proportion of REF reads that have secondary alignments 0
altsecondprop Proportion of ALT reads that have secondary alignments 0
refmatecontigcount Number of contigs seen in REF reads 1
altmatecontigcount Number of contigs seen in ALT reads 1
sixtypes Types of SNV (of the six possible types) C>T/G>A

## Project details

Uploaded source
Uploaded py3