Fork me on GitHub!

Snakemake workflow for TITAN

Description

This workflow will run the TITAN a set of tumour-normal pairs, starting from the BAM files and generating TitanCNA outputs. It will also perform model selection at the end of the workflow to choose the optimal ploidy and clonal cluster solutions.

Contact

Gavin Ha
Fred Hutchinson Cancer Research Center
contact: gavinha@gmail.com or gha@fredhutch.org
Date: May 11, 2019
Website: GavinHaLab.org

Requirements

Software packages or libraries

  • R-3.5
    • TitanCNA (v1.15.0+) - TitanCNA imports: GenomicRanges, dplyr, data.table, doMC
    • ichorCNA (v0.1.0)
    • HMMcopy
    • optparse
    • stringr
    • SNPchip
  • Python 3.4
    • snakemake-3.12.0
    • PySAM-0.11.2.1
    • PyYAML-3.12
  • samtools v1.3.1
  • bcftools v1.1
  • HMMcopy Suite.
    -In particular, readCounter is used.

Scripts/executables

  • readCounter (C++ executable; HMMcopy Suite)
  • ichorCNA.R (ichorCNA tool for normalizing and correcting read coverage)
  • countPysam.py (generates input allele counts)
  • titanCNA.R (main R script to run TitanCNA)
  • selectSolution.R (R script to select optimal solution for each sample)

Tumour-Normal sample list

The list of tumour-normal paired samples should be defined in a YAML file. See config/samples.yaml for an example. Both fields samples and pairings must to be provided. pairings key must match the tumour sample while the value must match the normal sample.

samples:
  tumor_sample_1:  /path/to/bam/tumor.bam
  normal_sample_1:  /path/to/bam/normal.bam


pairings:
  tumor_sample_1:  normal_sample_1

snakefiles

  1. ichorCNA.snakefile
  2. getAlleleCounts.snakefile
  3. TitanCNA.snakefile

config.yaml

See below for details about config/config.yaml

Run the analysis

1. Invoking the full snakemake workflow on a single machine

This will also run both ichorCNA.snakefile and getAlleleCounts.snakefile which generate the necessary inputs for TitanCNA.snakfile.

# show commands and workflow
snakemake -s TitanCNA.snakefile -np
# run the workflow locally using 5 cores
snakemake -s TitanCNA.snakefile --cores 5

2. Invoking the snakemake workflow on a cluster

Here are instructions for running workflow on a cluster using specific resource settings for memory and runtime limits, and parallel environments.
There are two cluster configurations provided: qsub and slurm

a. qsub

There are 2 separate files in use for qsub, which are provided as a template: config/cluster_qsub.sh - This file contains other qsub parameters. Note that these settings are used for the Broad’s UGER cluster so users will need to modify this for their own clusters.
config/cluster_qsub.yaml - This file contains the memory, runtime, and number of cores for certain tasks.

To invoke the snakemake pipeline for qsub:

snakemake -s TitanCNA.snakefile --jobscript config/cluster_qsub.sh --cluster-config config/cluster_qsub.yaml --cluster-sync "qsub -l h_vmem={cluster.h_vmem},h_rt={cluster.h_rt} -pe {cluster.pe} -binding {cluster.binding}" -j 50

Here, the h_vmem (max memory), h_rt (max runtime) are used. For runTitanCNA task, the default setting is to use 1 core but additional number of cpus (per task) can help to speed up the analysis. This can be set with -pe and -binding. Your SGE settings may be different and users should adjust accordingly.

b. slurm

There is only one file in use for slurm: config/cluster_slurm.yaml - This file contains the memory, runtime, and number of cores for certain tasks. To invoke the snakemake pipeline for qsub:

snakemake -s TitanCNA.snakefile --cluster-config config/cluster_slurm.yaml --cluster "sbatch -p {cluster.partition} --mem={cluster.mem} -t {cluster.time} -c {cluster.ncpus} -n {cluster.ntasks} -o {cluster.output}" -j 50

3. Invoking individual steps in the workflow

Users can run the snakemake files individually. This can be helpful for testing each step or if you only wish to generate results for a particular step. The snakefiles need to be run in this same order since input files are generated by the previous steps. ### a. ichorCNA.snakefile i. Compute read counts for non-overlapping bins of a specified size. This step uses HMMcopy’s readCounter ii. Run ichorCNA analysis to generate normalized read coverage-based copy number. For male samples, chrX results will be used from this step.

  snakemake -s ichorCNA.snakefile -np
  snakemake -s ichorCNA.snakefile --cores 5
  # OR
  snakemake -s ichorCNA.snakefile --jobscript config/cluster_qsub.sh --cluster-config config/cluster_qsub.yaml --cluster-sync "qsub -l h_vmem={cluster.h_vmem},h_rt={cluster.h_rt} -pe {cluster.pe} -binding {cluster.binding}" -j 50
  # OR
  snakemake -s ichorCNA.snakefile --cluster-config config/cluster_slurm.yaml --cluster "sbatch -p {cluster.partition} --mem={cluster.mem} -t {cluster.time} -c {cluster.ncpus} -n {cluster.ntasks} -o {cluster.output}" -j 50

Note: This snakefile for ichorCNA is specific for TitanCNA and may differ from the snakemake workflow on https://github.com/broadinstitute/ichorCNA

### b. getAlleleCounts.snakefile i. Applies samtools mpileup command to extract heterozygous SNP sites. The sites that overlap a SNP database are kept. Here is an example of a database (VCF format) that users can use: hapmap_3.3.hg38.vcf.gz. ii. Extract the allelic read counts from the tumor bam file, parallelized by each chromosome.
iii. Combine the allelic read counts from each chromosome file into a single counts file.

  snakemake -s getAlleleCounts.snakefile -np
  snakemake -s getAlleleCounts.snakefile --cores 5
  # OR
  snakemake -s getAlleleCounts.snakefile --jobscript config/cluster_qsub.sh --cluster-config config/cluster_qsub.yaml --cluster-sync "qsub -l h_vmem={cluster.h_vmem},h_rt={cluster.h_rt} -pe {cluster.pe} -binding {cluster.binding}" -j 50
  # OR
  snakemake -s getAlleleCounts.snakefile --cluster-config config/cluster_slurm.yaml --cluster "sbatch -p {cluster.partition} --mem={cluster.mem} -t {cluster.time} -c {cluster.ncpus} -n {cluster.ntasks} -o {cluster.output}" -j 50

### c. TitanCNA.snakefile i. Run the TitanCNA analysis and generates solutions for different ploidy initializations and each clonal cluster.
ii. Merge results with ichorCNA output generate by ichorCNA.snakefile and post-processes copy number results. In particular, it combines chrX results from ichorCNA for male samples. iii. Select optimal solution for each samples and copies these to a new folder. The parameters are compiled in a text file.
iv. Creates a new directory containing symbolic links to result files for optimal solutions of each sample.

Configuration and settings

All settings for the workflow are contained in config/config.yaml. The settings are organized by paths to scripts and reference files and then by each step in the workflow.

1. Path to tools

samTools is used by (getAlleleCounts.snakefile)[getAlleleCounts.snakefile].

samTools:  /path/to/samtools

2. Path to scripts

These are provided in this (TitanCNA)[https://github.com/gavinha/TitanCNA] repo either under code/ or R_scripts.

readCounterScript:  /path/to/readCounter
ichorCNA_rscript:  /path/to/ichorCNA.R
pyCountScript:  code/countPysam.py
TitanCNA_rscript: ../R_scripts/titanCNA.R
TitanCNA_combineTitanIchorCNA:  code/combineTITAN-ichor.R
TitanCNA_selectSolutionRscript: ../R_scripts/selectSolution.R

See the ichorCNA repo for the ichorCNA R script.

3. Path to R package files

Specify the directory in which TitanCNA and ichorCNA are installed.
Set these if the R files in these libraries have been modified or updated but not yet installed or updated in R.

TitanCNA_libdir:  ../../R/
ichorCNA_libdir:  /path/to/ichorCNA/ ## optional

4. Reference files and settings

Global reference files used by many of the snakefiles and scripts.

  • snpVCF you can download the HapMap file (used for filtering heterozygous SNPs) here: https://console.cloud.google.com/storage/browser/genomics-public-data/resources/broad/hg38/v0
  • genomeStyle specifies the chromosome naming convention to used for output files. Input files can be any convention as long as it is the same genome build. Only use UCSC (e.g. chr1) or NCBI (e.g. 1).
  • sex set to male or female, otherwise None if both females and males are in sample set.
  • cytobandFile Only used for hg38! Coordinates for plotting the idiogram. The file is found in the (ichorCNA repo)[https://github.com/broadinstitute/ichorCNA/blob/master/inst/extdata/cytoBand_hg38.txt].
    genomeBuild: hg38 # Use "None" if hg19
    genomeStyle:  UCSC
    refFasta: /path/to/Homo_sapiens_assembly38.fasta
    snpVCF:  /path/to/hapmap_3.3.hg38.vcf.gz 
    cytobandFile:  data/cytoBand_hg38.txt # only need if hg38
    centromere:  /path/to/ichorCNA/inst/extdata/GRCh38.GCA_000001405.2_centromere_acen.txt
    sex:  male   # use "None" if both females and males are in sample set
    

5. readCounter settings

Settings for the computing read coverage using readCounter.

  • chrs Used by readCounter command line tool. Please comment out the chromosome naming convention that you do not wish to use.
  • binSize Various bin sizes can be used in the analysis.
    ## readCounter params ##
    # use this for NCBI chr naming
    chrs: 1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,X,Y 
    # use this for UCSC chr naming
    #chrs: chr1,chr2,chr3,chr4,chr5,chr6,chr7,chr8,chr9,chr10,chr11,chr12,chr13,chr14,chr15,chr16,chr17,chr18,chr19,chr20,chr21,chr22,chrX,chrY 
    binSize:  10000
    

6. ichorCNA.snakefile settings

Settings for the analysis of read coverage using ichorCNA.

  • ichorCNA_chrs specifies the chromosomes to analyze in the R script; users do not need to be concerned about chromosome naming convention here as the code will handle it based on the genomeStyle set in the reference settings above.
  • The GC and Map wig files must have match the binSize above. The various sizes supported will depend on which GC and Map wig files are generated. We provide a few to select from in the ichorCNA repo. Users can also create their own by following instructions from the HMMcopy repo.
    ## ichorCNA params ##
    ichorCNA_gcWig: /path/to/gc_hg19_10kb.wig
    ichorCNA_mapWig:  /path/to/map_hg19_10kb.wig
    ichorCNA_chrs:  c(1:22, \"X\")
    ichorCNA_normal:  c(0.5)  
    ichorCNA_ploidy:  c(2,3)  
    ichorCNA_estimateNormal:  TRUE
    ichorCNA_estimatePloidy:  TRUE
    ichorCNA_estimateClonality: TRUE
    ichorCNA_scStates:  c(1,3)
    ichorCNA_maxCN:  8
    ichorCNA_includeHOMD: FALSE
    ichorCNA_txnE:  0.9999
    ichorCNA_txnStrength:  10000
    ichorCNA_plotFileType:  png
    ichorCNA_plotYlim:  c(-2,4) 
    

7. getAlleleCounts.snakefile settings: Tumor allelic counts

Minimum thresholds used when extracting read counts from the tumor BAM file at heterozygous SNP sites. In addition, the vcf_quality of the heterozygous SNP site from the matched normal sample is also used in the filtering. Note: Users must modify getAlleleCounts.snakefile to set the appropriate chromosome naming convention for samtools and the countPysam.py script. Comment out line 6 if using UCSC or comment out line 8 when using NCBI.

# USERS MUST MODIFY getAlleleCounts.snakefile to use the correct CHRS naming
map_quality:  10
base_quality: 10
vcf_quality:  100

8. TitanCNA.snakefile settings

Most settings can be left as default.

  • TitanCNA_maxNumClonalClusters specifies the maximum number of clonal clusters to consider. For example, if set to 5, then 5 solutions are generated, each one considering a different number of cluster(s).
  • TitanCNA_maxPloidy specifies the maximum ploidy to initialize. This be set to either 2 (only considers diploid solutions), 3 (considers diploid and triploid, and usually accounts for tetraploid), or 4 (for diploid, triploid, tetraploid or higher ploidies). Usually, 3 is suitable for most samples unless you know that your samples are tetraploid or even higher. For example, if set to 3, then solutions for diploid and triploid will be generated. code/selectSolution.R will try to select the optimal solution; however, users should inspect to make sure results are accurate.
  • TitanCNA_numCores specifies the number of cores to use on a single machine. Ff using a cluster, then must match the settings for number of cpus in config/cluster_qsub.yaml or config/cluster_slurm.yaml.
    TitanCNA_maxNumClonalClusters: 2
    TitanCNA_chrs:  c(1:22, \"X\")
    TitanCNA_normalInit: 0.5
    TitanCNA_maxPloidy: 3
    TitanCNA_estimateNormal:  map
    TitanCNA_estimatePloidy:  TRUE
    TitanCNA_estimateClonality: TRUE
    TitanCNA_alleleModel: binomial
    TitanCNA_alphaK:  10000
    TitanCNA_alphaR:  10000
    TitanCNA_txnExpLen: 1e15
    TitanCNA_plotYlim:  c(-2,4)
    TitanCNA_solutionThreshold: 0.05
    TitanCNA_numCores: 1