Structural Variant Calling with Cortex-Var on Nextflow

This repo is a variant calling pipeline that uses cortex_var written in Nextflow, for use in detecting de-novo structural variance in whole-genome reads.

Many information mentioned in this readme are referred from the cortex_var manual

You may use this README file to understand how the scripts are organized and how the codes are structured. Additionally, this README file also provides a brief explanation about how cortex-var behaves.

Files in this repo are organized as follows:

Folder / Files	Contents
runCortexVarHighCoverage	Nextflow script to be run by user
runCortexVarLowCoverage	Nextflow script to be run by user
nextflow.config	Configuration file to be edited by user
nextflow_scripts	Individual nextflow scripts for each step in cortex-var
usefulCalculators	Tools to help finding some parameters in the config file

Table of Contents

Structural Variant Calling with Cortex-Var on Nextflow
- Intended pipeline architecture and function
  - Installation
- User Guide

Intended pipeline architecture and function

This workflow implements Cortex-var workflow for structural variant calling in whole genome reads.

Cortex suggests to use high coverage reads, as suggested by section 9.2 of the cortex_var manual, using the principle of ("things which happen rarely are more likely to be errors than real"), and according section 6.2 of the cortex_var manual, the threshold considered as "high" or "low" coverage samples is about above or below 20x coverage for diploid and 10x for above or below 10x for haploid.

The standard pipeline for cortex using high coverage is as such:

Create de Bruijn graph for each sample (paired reads)
Clean individual sample de Bruijn graph
Create reference de Bruijn graph
Create multi-color graph, consisting of reference de Bruijn graph and cleaned sample de Bruijn graphs
Call variants by Bubble Caller and/or Path Divergence
Convert cortex calls to VCF

Figure 1: Cortex_var High-Coverage Pipeline Overview

Additionally, cortex_var also supports a pipeline using many low coverage samples from the same population, also suggested by section 9.2 of the cortex_var manual.

Create de Bruijn graph for each sample
Create pooled de Bruijn graph
Clean individual sample de Bruijn graph based on pooled de Bruijn graph
Create reference de Bruijn graph
Create multi-color graph, consisting of reference de Bruijn graph and cleaned sample de Bruijn graphs
Call variants by Bubble Caller and/or Path Divergence
Convert cortex calls to VCF

Figure 2: Cortex_var Low-Coverage Pipeline Overview

IMPORTANT NOTE: Cortex var works chronologically. Which means to conduct one process, the previous processes must be done first. Although it is possible to do each step separately, the output of previous steps must still be present as an input for the next step. For a clear visual which process requires which process input, please refer to diagrams above. An arrow from process a to process b indicates that process b requires all files related to process a (both input and output) to still be present for process b to run successfully.

Installation

Dependencies

cortex_var
Nextflow 0.30.1.4844 or newer is recommeded

Cortex_var installation

Refer to cortex_var github page

Workflow Installation

Clone this repository

Modifying process_calls.pl

To run conversion VCF, it is needed to change the perl script process_calls.pl within /cortex/scripts/analyse_variants in the cortex directory. Lines 2273 - 2283 should be changed into:

if ( $two_alleles =~ /^([ACGTRYKMSWBDHVN]+)_([ACGTRYKMSWBDHVN]+)$/ )
                       {
                               $this_indel_ref_allele = $1;
                               $this_indel_alt_allele = $2;
                       }
                       elsif ( $two_alleles =~ /^_([ACGTRYKMSWBDHVN]+)$/ )
                       {
                               $this_indel_ref_allele = "";
                               $this_indel_alt_allele = $1;
                       }
                       elsif ( $two_alleles =~ /^([ACGTRYKMSWBDHVN]+)_$/ )

Creating symlink for perl folder in vcftools

Cortex_var was created before the /perl folder in vcftools were gone. So when the vcftools install is new, cortex_var cannot convert cortex calls to vcf files, as it is looking for /perl/vcf-sort within the vcftools folder. Thus, it is necessary to create a symlink in the vcftools variable specified in nextflow.config file that has the name /perl but leads to the /bin directory of vcftools.

User Guide

Data Preparation

Cortex_var requires two types of data input:

Sample pair reads
Reference fasta files, separated by chromosome

Sample Pair Reads

Sample pair reads are required to be placed in the same folder, and the path to this folder will later be the input for sampleDir parameter in nextflow.config.

Figure 3: sampleDir Example

Reference Fasta Files

Cortex requires the user to input a list of path to the reference fasta files, separated by chromosomes to make reference de Bruijn graphs and Path Divergence variant calling. The reference fasta files of each chromosome should be listed as paths, separated by line break (\n) that will have its path specified in pathToReferenceList parameter in nextflow.config. Example of what the textfile should look like is as follows:

Figure 4: Example of Reference Fasta List Textfile

Cortex_Var Executable Preparation

This workflow requires the user to have some prior knowledge about cortex_var, so this section will be brief. Cortex_var requires the user to have cortex_var executable already made, specific to color and k-mer. To do this, the user can refer to the INSTALL file in the cortex_var repository.

"Color" in cortex simply means how many de Bruijn Graph files (.ctx extensions) will be loaded to the cortex executable. A simple way of thinking about "color" is how many genomes are loaded in the executable to be processed.

For each step, here is the guide to how many colors is needed: Note: n is the number of samples (1 sample requires 2 reads). For example: if there are 2 samples there would be 4 reads, and the value of n is 2.

For standard, high coverage samples:

makeSampleGraph: 1 color
cleanGraphPerSampleHighCoverage: 1 color
makeReferenceGraph: 1 color
makeCombinationGraph: finalCombinationGraphMaxColor
variantCalling: finalCombinationGraphMaxColor

For low coverage, pooled base pipeline:

makeSampleGraph: 1 color
poolAndCleanErrors: 1 color
cleanGraphPerSampleLowCoverage: 2 colors
makeReferenceGraph: 1 color
makeCombinationGraph: finalCombinationGraphMaxColor
variantCalling: finalCombinationGraphMaxColor

"Kmer size" in cortex dictates the number of bases to be used to construct the node in the de Bruijn graph made. The Kmer size is required to be consistent throughout the process.

Cortex_var also requires a specific kmer size to be specified by user. To choose the appropriate kmer size, refer to section 8 of the cortex_var manual.

After choosing the kmer and color for each cortex executable each of them much be made and compiled in the cortex bin directory specified, refer to section 4 of cortex_var manual.

For example, if color chosen is 1 and kmer size chosen is 63:

cd to installed cortex_var directory (bash install.sh has already been run)
make NUM_COLS=1 MAXK=63 cortex_var

After running the "make" commands, within the cortexDir/bin directory, the executable cortex_var_63_c1 will be made, "63" being the kmer size, and "c1" indicates the number of colors.

finalCombinationGraphMaxColor is a parameter to be filled by the user, this indicates the color required by makeCombinationGraph and variantCalling step. Further explanation on how to fill this parameter will be provided in the Nextflow.config Parameters section below.

Nextflow config Parameters

The workflow is controlled by modifying nextflow.config file. Note: String parameters always need quotation marks ("") in the beginning and end of the string.

Executables

nextflowDir - STRING

Path to nextflow executable

cortexBinDir - STRING

Path to bin directory of cortex, where executables for each process is located.

Sample, Result, and Config Directory

sampleDir - STRING

Path to directory where samples are located.

resultsDir - STRING

Path to directory where results of each process will be dumped.

logDir - STRING

Path to directory where logs of each process will be dumped, default is inside the resultsDir

cloneDir - STRING

Path to where this repository is cloned (path/to/GenomicsCortexVarNextflow)

Sample Management

All samples are organized as referred to figure 2, the fastq filenames should be organized as: "(sampleName)(sampleReadPattern)(readNumber)(sampleReadExtension)"

For instance, for the following samples: sampleDir/sample1_read1.fq.gz sampleDir/sample1_read2.fq.gz sampleDir/sample2_read1.fq.gz sampleDir/sample2_read2.fq.gz

sampleName is sample1 and sample2 sampleReadPattern is _read readNumber is 1 and 2 for each sample sampleReadExtension is .fq.gz

sampleList - ARRAYLIST OF STRING

ArrayList containing the sample names enclosed within quotation marks and separated by comma, all enclosed within []

With examples given, the sampleList should be filled with: ["Sample_1", "Sample_2"]

Note: Quotation marks are required to indicate that the element is a string.

sampleReadPattern - STRING

The string after the sample name, but before the read number.

With examples given, the sampleReadPattern should be filled with: "_read"

sampleReadExtension - STRING

The string after the read number, cortex supports both raw .fq and gzipped .fq.gz

In example given, sampleReadExtension should be filled with ".fq.gz"

pathToReferenceList - STRING

Path to where the file containing reference fasta files paths. Example: "/PATH/TO/REFERENCELIST.TXT".

Refer to Reference Fasta Files

Executor

executor - STRING

The type of executor that will be used by user. refer to nextflow executors documentation

Cortex_var Individual Process Parameters

Generic process parameters

note: excluding conversion to VCF

kmerSize - STRING

kmerSize chosen for the whole cortex_var process, refer to [Cortex_Var Executable Preparation](#Cortex_Var Executable Preparation)

cortexConfig - STRING

Cortex_var requires flags --kmer_size --mem_height and --mem_width for each step. kmer_size indicates the length of each k-mer (nodes in the graph) and is required to be consistent with the cortex binary executables.

By default, the kmer size will be derived from kmerSize parameter above, but the memory height and width should be specified here.

mem_height indicates the height of the hash table, and mem_width indicates the width of hash table. If the user needs help in determining the mem_height and mem_width, executing /usefulCalculators/findHeightAndWidth.py can be of assistance. This script is written in python3 and will take in the size of genome in number of bases and returns the optimum mem_height and mem_width, according to the cortex_var manual page 8.

For example, soybean genome has about 1.15 gigabases:

#"Enter genome size in bases:" will be prompted when running the script.
Enter genome size in bases: 1150000000 
#Outputs below
mem_height = 27
mem_width = 18

If the kmer_size is 63 combined with the example above, the input for cortexConfig(PROCESSNAME) should be: "--kmer_size " + kmerSize + " --mem_height 27 --mem_width 18"

run(PROCESSNAME) - STRING

Indicate if user is willing to run that specific process. ("y"/"n")

cortexBin(PROCESSNAME) - STRING

Executable to where the executable will be used for the process. Default is in the bin directory specificied by cortexBinDir

(PROCESSNAME)Queue - STRING

Queue name in the cluster, refer to nextflow executor documentation

(PROCESSNAME)MaxNodes - INT

Maximum number of nodes to be used in parallel, refer to nextflow maxForks documentation

(PROCESSNAME)Walltime - STRING

Walltime for individual process runs, refer to nextflow walltime documentation

(PROCESSNAME)CpusNeeded - STRING

Number of cores per node for each process, refer to nextflow cpus documentation

(PROCESSNAME)ErrorStrategy - STRING

For what is the action taken when process encounters error(s) documentation

Specific process parameters

Make sample de Bruijn graph specific parameters

quality_score_threshold - INT

Initial quality filter for making de Bruijn graph, refer to section 6.2 and page 3 of cortex_var manual

Make combination graph specific parameters

pathToRefCtx - STRING

Path to reference de Bruijn graph, usually having .ctx extension (default is path to where it would be made by makeReferenceGraph)

finalCombinationGraphMaxColor - INT

The largest number of colors of graph to run the node can handle, given kmer size, memory height and memory width.

This process creates multi-colored deBruijn graph, stacking reference de Bruijn graph with one or more cleaned sample de Bruijn graph(s). Thus, the number ranges from 2 to N+1 (total number of samples), depending on the memory capability of the node.

To identify how much memory the process needs, the user can refer to cortex_var manual page 8. However, this repository also provides a script to calculate memory required, /usefulCalculators/findMemoryRequired.py. For more information, refer to Resource Requirement section [below](#Resource Requirement)

If the finalCombinationGraphMaxColor is not N+1, then this pipeline will automatically parition the graphs, and make the total of N %(modulo) (finalCombinationGraphMaxColor - 1) combination graphs.

Variant Calling specific parameters

PD - STRING

Indicate if user wants to run Path Divergence variant calling method ("y"/"n")

For more information regarding Path Divergence variant calling method, refer to Section 11 of [cortex_var manual] (http://cortexassembler.sourceforge.net/cortex_var_user_manual.pdf)

Parameters used only if PD is selected

populationPD - STRING

Indicate if user wants to run combined Path Divergence variant calling in the same combination graph consecutively ("y"/"n")

After the vcf conversion process, variant calls of every sample in the same combination graph will be available in a single VCF file, one per combination graph.

For more information, refer to page 12, or section 11.2 of [cortex_var manual] (http://cortexassembler.sourceforge.net/cortex_var_user_manual.pdf)

individualPD - STRING

Indicate if user wants to run path divergence on one of one sample for each process ("y"/"n")

After the vcf conversion process, variant calls of each individual sample will be available in individual vcf files.

For more information refer to page 12, or section 11.1 of [cortex_var manual] (http://cortexassembler.sourceforge.net/cortex_var_user_manual.pdf)

maxVarLength - INT

Maximum variant length to look for, default is 40000 base pairs

For more information refer to page 18 of [cortex_var manual] (http://cortexassembler.sourceforge.net/cortex_var_user_manual.pdf)

VCF Caller specific parameters

stampyBin - STRING

Full path to stampy.py.

For more information refer to page 15 of [cortex_var manual] (http://cortexassembler.sourceforge.net/cortex_var_user_manual.pdf)

stampyHashProducts - STRING

For more information refer to page 15 of [cortex_var manual] (http://cortexassembler.sourceforge.net/cortex_var_user_manual.pdf)

VCFToolsDir - STRING

For more information refer to page 15 of [cortex_var manual] (http://cortexassembler.sourceforge.net/cortex_var_user_manual.pdf)

samplePloidy - INT

Ploidy of sample.

Example: if sample is diploid, samplePloidy should be 2

referenceFasta

Path to reference fasta file. Note: this is different from pathToReferenceList.

For more information refer to page 15 of [cortex_var manual] (http://cortexassembler.sourceforge.net/cortex_var_user_manual.pdf)

Resource Requirements

The main resources cortex_var needs are memory and time. The amount of memory cortex_var require depends on the size of hash table specified in cortexConfig(PROCESSNAME) parameters, color number, and kmer size. To compute the approximate ram required, the python3 script usefulCalculators/findMemoryRequired.py is available.

Example:

enter kmer size: 63
Enter height: 27
Enter width: 18
enter number of colors: 2
Memory Required in GB: 77.309411328
Memory Required in TB: 0.077309411328
FOR IFORGE USERS ONLY: Use big_mem

With this example, this means that the user should fill (PROCESSNAME)Queue with a queue that have at least 77.3 GB of RAM (more is recommended).

Executing Nextflow Application

The suggested practice to execute nextflow is to place the folder containing binary executable for nextflow in the PATH environment or to use full path to the nextflow executable file.

There are two master scripts available: runCortexVarHighCoverage.nf and runCortexVarLowCoverage.nf, as its respective names suggest, if the sample read coverage is high, running the standard pipeline nextflow run runCortexVarHighCoverage will be appropriate. Respectively, with low sample read coverage, but coming from similar population, running low-coverage pipeline nextflow run runCortexVarLowCoverage.nf will be more appropriate. To determine what is a 'high' or 'low' coverage samples, the user can refer to section 6.2 of cortex_var manual .

Nextflow will create a work folder, where all logs and temporary files are kept (refer to [logging functionality](# Logging Functionality), therefore it is suggested (not mandatory) to avoid running the master control scripts runCortexVarHighCoverage.nf and/or runCortexVarLowCoverage.nf in some other directory, like the resultsDir, to keep the script directory clean.

Logging Functionality

Nextflow creates work folder where the command nextflow (script.nf) is run. Within the work folder, there will be nested folders, containing the log of individual processes. In the stdout of running script, there will be lines that is similar to this:

[73/a3f6d2] Submitted process > test

test is the process name. [73/a3f6d2] indicates the folder name within work folder when the user can locate the log for the specific process. More accurately, within /work/73/ there might be several folders, and a3f6d2 is the first 6 characters of the nested folder within /work/73/. Inside the folder specific to the process, such as /work/73/a3f6d2..., there are several log files that can be used for debugging or troubleshooting.

The work folder for each specific process is located within the resultsDir specified in the nextflow.config parameter.

Output Structure

This workflow places the output of each process described in figure 1 and figure 2 in the results directory specified in resultsDir parameter in nextflow.config file. Each process will have a folder, named as (PROCESSNAME)Output to host the output files of the specific process.

Test Environment

Executor : PBS
Nextflow Version : 30.1.4844
Genome source : Soybean
Read Sizes : 90bp - 100bp
Read Coverage: ~17x

ncsa/GenomicsCortextVarNextflow

Structural Variant Calling with Cortex-Var on Nextflow

Intended pipeline architecture and function

Installation

Dependencies

Cortex_var installation

Workflow Installation

Modifying process_calls.pl

Creating symlink for perl folder in vcftools

User Guide

Data Preparation

Sample Pair Reads

Reference Fasta Files

Cortex_Var Executable Preparation

Nextflow config Parameters

Executables

Sample, Result, and Config Directory

Sample Management

Executor

Cortex_var Individual Process Parameters

Generic process parameters

Specific process parameters

Make sample de Bruijn graph specific parameters

Make combination graph specific parameters

Variant Calling specific parameters

VCF Caller specific parameters

Resource Requirements

Executing Nextflow Application

Logging Functionality

Output Structure

Test Environment