intSiteCaller

---

Introduction

This code is designed to take fastq files that are produced by the MiSeq and return integration sites, multihits, and chimeras. RData and fasta files are used for data persistance in place of a database, thus allowing massive parallelization and the LSF job submission system on the PMACS HPC

Inputs

Analysis is started by having the user create the following directory structure:

primaryAnalysisDirectory
├── Data
│   ├── Undetermined_S0_L001_I1_001.fastq.gz
│   ├── Undetermined_S0_L001_R1_001.fastq.gz
│   └── Undetermined_S0_L001_R2_001.fastq.gz
├── processingParams.csv
└── sampleInfo.csv

Primary Analysis Directory

• Data/Undetermined_S0_L001_*_001.fastq.gz are the fastq files returned by the MiSeq (R1, R2, and I1)

• processingParams.csv contains 'dryside' sample metadata:

  • alias is the human-readable sample description
  • qualityThreshold, badQualityBases, qualitySlidingWindow
    • After error-correcting and demultiplexing, intSiteCaller trims raw MiSeq reads based on Illumina-issued quality scores. badQualityBases is the number of bases below qualityThreshold, using standard Illumina ASCII Q-score encoding (p.41-2), that can be observed in a window of qualitySlidingWindow before the read is trimmed.
  • primer is the primer sequence as seen in MiSeq read 2
  • ltrBit is the LTR sequence as seen in MiSeq read 2
  • largeLTRFrag is 43nt of the LTR sequence as seen from MiSeq read 1
  • linkerCommon is 15nt of the linker sequence as seen from MiSeq read 2
  • mingDNA is the minimum length of genomic DNA (in nt) that is allowed to be passed onto the alignment step
  • vectorSeq is a filepath (either absolute or relative to the primary analysis directory) to the vector sequence in fasta format -- it is encouraged to place the vector sequence directly in the primary analysis directory, although that is not a requirement
  • minPctIdent is the minimum percent identity that a query sequence has to a putative alignment on the target sequence in order to be considered 'valid'
  • maxAlignStart is the maximum number of nucleotides of the query sequence that do not match the putative target sequence before a matched nucleotide is seen
  • maxFragLength is the maximum length of a properly-paired alignment (in nt)
  • refGenome is the reference genome to be used for alignment - this is passed in as a standard text string (ex. 'hg18', 'hg19', 'mm8', 'mm9')

• sampleInfo.csv contains 'wetside' sample metadata:

  • alias is the human-readable sample description
  • linkerSequence is the linker sequence as seen in MiSeq read 1. N's indicate the presence of a primerID
  • bcSeq is the barcode used during sample preparation
  • gender is either 'M' of 'F' for male/female, respectively

Usage

After creating the above directory structure, the following command is issued:

Rscript intSiteCaller.R

The rest of the processing is fully automated and shouldn't take more than 4 hours to process 1.5e7 raw reads.

intSiteCaller.R can handle the following options

• -j, --jobID - Unique name by which to identify this intance of intSiteCaller [default: intSiteCallerJob]
• -c, --codeDir - Directory where intSiteCaller code is stored, can be relative or absolute [default: .]
• -p, --primaryAnalysisDir - Location of primary analysis directory, can be relative or absolute [default: .]
• -C, --cleanup - Remove temporary files upon successful execution of intSiteCaller
• -h, --help - Show the help message and exit

Outputs

Integration sites

This code returns integration sites in two formats. allSites.RData is a GRanges object that contains a single record for each Illumina read. sites.final.RData is a GRanges object of dereplicated integration sites along with a tally of how many reads were seen for each site (irregardless of sonic breakpoint). The revmap column in sites.final.RData links unique reads from allSites to dereplicated sites in sites.final.

Multihits

Multihits are stored in multihitData.RData which is a list of three items:

• unclusteredMultihits is a GRanges object of raw multihits, directly from properlyPairedAlignments
• clusteredMultihitPositions is a GRangesList object of clustered solostart multihits with each included GRanges object representing one multihit cluster
• clusteredMultihitLengths is a list object containing dataframes of multihit 'fragment lengths' for use in sonicAbundance. Each list record, n, cooresponds to the nth cluster as defined in clusteredMultihitPositions. The Var1 column cooresponds to the 'length' and the Freq column cooresponds to the number of times that fragment length was observed.

Chimeras

Chimeras are stored in chimeraData.RData which is a list that contains some basic chimera frequency statistics and a GRangesList object. Each GRanges object contains two records, one for the read1 alignment and another for the read2 alignment

PrimerIDs

PrimerIDs (if present in the linker sequence) are stored in primerIDData.RData. This file is a base R list containing a DNAStringSet and a BStringSet containing the sequences and quality scores of the primerIDs.

Stats

Processing statistics are returned in the stats.RData file. This file contains a single data.frame. Detail of the specific columns provided in this dataframe will be added later and can inferred from intSiteLogic.R.

Dependencies

This code is highly dependent on Bioconductor packages for processing DNA data and collapsing/expanding alignments.

The following R packages and their subsesequent dependencies are required for proper operation of intSiteCaller:

• ShortRead
• hiReadsProcessor
• GenomicRanges
• rtracklayer
• BSgenome
• argparse
• igraph
• Any BSgenome.*.UCSC.* package cooresponding to reference genomes specified in processingParams.csv

Specific versioning analysis has not yet been performed.

Additionally, BLAT code requires the availability of the blat and python command. blat is available on PMACS at /opt/software/blatSrc/v35/bin/x86_64, however is not included in the default PATH. This directory will need to be added to the user's default PATH in order to successfullly run BLAT alignments.

intSiteCaller confirms the presence of all dependancies and will throw an error if a dependancy is not met.

Code Structure

• Primary read trimming and integration site calling logic is contained in intSiteLogic.R.
• Branching and condensing of PMACS jobs is performed in programFlow.R
• Barcode error correcting logic is performed in errorCorrectIndices/golay.py as wrapped by errorCorrectIndices/processGolay.py.
• All code files are checked into the repository.
• Flowcharts will be added to graphically describe the flow of the overall program as well as the flow/logic of individual functions

Tests

A sample dataset is included for verification of integration site calling accuracy. The testCases directory contains a subdirectory, intSiteValidation, and a compressed folder, intSiteValidationOUTPUT.tar.gz. To analyze the test data, move intSiteValidation/* to an LSF envrionment and execute the code as described in the 'Usage' section. Upon successful execution, the directory should resemble intSiteValidationOUTPUT.tar.gz. Note that this subset of data contains samples with some borderline cases. For example, clone7 samples should all fail, and many of the clone1-clone4 samples should return no multihits or chimeras. The current implementation of the code handles these gracefully.