# Overview of the recab function of bamUtil

The recab option of bamUtil recalibrates a SAM/BAM file.

Recalibration can also be called as an option of bamUtil: dedup. This will perform the recalibration and the deduping in the same set of steps, increasing processing speed.

## Handling Recalibration/Implementation Notes

Recalibration is a 2-step process that loops through the file twice (stdin is not support as input):

1. Build Recalibration Table
2. Apply Recalibration Table

The Recalibration Table groups bases based on a set of covariates:

• Quality (either from the quality string or from a tag)
• Cycle (reverse complement for reverse strands)
• 1st/2nd read in pair
• Previous Cycle's Base (reverse complement for reverse strands)
• This Cycle's Base (reverse complement for reverse strands)

The Recalibration Table tracks the number of matches/mismatches for each set of covariates.

Only bases meeting all of the following criteria are used to Build the Recalibration Table:

The Recalibration Table is applied to all bases meeting all of the following criteria (even if they were not used for creating the table):

Recalibrated Quality is: $-10 * \log \frac{mismatches + 1}{mismatches + matches + 1}$

Alternatively, logistic regression can be used for calculating the new quality.

If the Recalibrated Quality is greater than maxBaseQual, the updated quality is set to maxBaseQual.

Optionally, the previous quality can be stored in a tag.

The current recalibration logic was designed for recalibrating ILLUMINA data.

NOTE: GATK ignores/skips adapters, but our logic does not.

## How to use it

When recab is invoked without any arguments the usage information is displayed as described below under Usage.

The input SAM/BAM file (--in), the output SAM/BAM file (--out), and the reference file (--refFile) are required inputs.

Recommended usage with Deduper:

/usr/cluster/bin/bam dedup --recab --in ${INPUT}.bam --out${OUTPUT}.bam --force --refFile ${REF} --dbsnp${DBSNP} --oneChrom --storeQualTag OQ --maxBaseQual 40


Recommended usage without Deduper:

/usr/cluster/bin/bam recab --in ${INPUT}.bam --out${OUTPUT}.bam --refFile ${REF} --dbsnp${DBSNP} --storeQualTag OQ --maxBaseQual 40


# Usage

./bam recab (options) --in <InputBamFile> --out <OutputFile> [--log <logFile>] [--verbose] [--noeof] [--params] --refFile <ReferenceFile> [--dbsnp <dbsnpFile>] [--minBaseQual <minBaseQual>] [--maxBaseQual <maxBaseQual>] [--blended <weight>] [--fitModel] [--fast] [--keepPrevDbsnp] [--keepPrevNonAdjacent] [--useLogReg] [--qualField <tag>] [--storeQualTag <tag>] [--buildExcludeFlags <flag>] [--applyExcludeFlags <flag>]


# Parameters

Required General Parameters :
--in <infile>   : input BAM file name
--out <outfile> : output recalibration file name
Optional General Parameters :
--log <logfile> : log and summary statistics (default: [outfile].log)
--verbose       : Turn on verbose mode
--noeof         : do not expect an EOF block on a bam file.
--params        : print the parameter settings

Recab Specific Required Parameters
--refFile <reference file>    : reference file name
Recab Specific Optional Parameters :
--dbsnp <known variance file> : dbsnp file of positions
--minBaseQual <minBaseQual>   : minimum base quality of bases to recalibrate (default: 5)
--maxBaseQual <maxBaseQual>   : maximum recalibrated base quality (default: 50)
qualities over this value will be set to this value.
This setting is applied after binning (if applicable).
--blended <weight>            : blended model weight
--fitModel                    : check if the logistic regression model fits the data
overriden by fast, but automatically applied by useLogReg
--fast                        : use a compact representation that only allows:
* at most 256 Read Groups
* maximum quality 63
* at most 127 cycles
overrides fitModel, but is overridden by useLogReg
uses up to about 2.25G more memory than running without --fast.
--keepPrevDbsnp               : do not exclude entries where the previous base is in dbsnp when
building the recalibration table
By default they are excluded from the table.
--keepPrevNonAdjacent         : do not exclude entries where the previous base is not adjacent
(not a Cigar M/X/=) when building the recalibration table
By default they are excluded from the table (except the first cycle).
--useLogReg                   : use logistic regression calculated quality for the new quality
automatically applies fitModel and overrides fast.
--qualField <quality tag>     : tag to get the starting base quality
(default is to get it from the Quality field)
--storeQualTag <quality tag>  : tag to store the previous quality into
--buildExcludeFlags <flag>    : exclude reads with any of these flags set when building the
recalibration table.  Default is 0xF04
--applyExcludeFlags <flag>    : do not apply the recalibration table to any reads with any of these flags set
Quality Binning Parameters (optional):
Bin qualities by phred score, into the ranges specified by binQualS or binQualF (both cannot be used)
Ranges are specified by comma separated minimum phred score for the bin, example: 1,17,20,30,40,50,70
The first bin always starts at 0, so does not need to be specified.
By default, the bin value is the low end of the range.
--binQualS   : Bin the Qualities as specified (phred): minQualOfBin2, minQualofBin3...
--binQualF   : Bin the Qualities based on the specified file
--binMid     : Use the mid point of the quality bin range for the quality value of the bin.
--binHigh    : Use the high end of the quality bin range for the quality value of the bin.


	PhoneHome:
--noPhoneHome       : disable PhoneHome (default enabled)
--phoneHomeThinning : adjust the PhoneHome thinning parameter (default 50)


## Required Generic Parameters

### Input File (--in)

Use --in followed by your file name to specify the SAM/BAM input file.

The program automatically determines if your input file is SAM/BAM/uncompressed BAM without any input other than a filename from the user.

Note: This tool does not support input from stdin.

 SAM/BAM/Uncompressed BAM from file --in yourFileName

Note: Uncompressed BAM is compressed using compression level-0 (so it is not an entirely uncompressed file). This matches the samtools implementation so pipes between our tools and samtools are supported.

### Output File (--out)

Use --out followed by your file name to specify the SAM/BAM output file.

The file extension is used to determine whether to write SAM/BAM/uncompressed BAM. A - is used to indicate stdout and the extension for file type (no extension is SAM).

 SAM to file --out yourFileName.sam BAM to file --out yourFileName.bam Uncompressed BAM to file --out yourFileName.ubam SAM to stdout --out - BAM to stdout --out -.bam Uncompressed BAM to stdout --out -.ubam

Note: Uncompressed BAM is compressed using compression level-0 (so it is not an entirely uncompressed file). This matches the samtools implementation so pipes between our tools and samtools are supported.

## Optional Generic Parameters

### Output log & Summary Statistics FileName (--log)

Output file name for writing logs & summary statistics.

If this parameter is not specified, it will write to the output file specified in --out + ".log". Or if the output bam is written to stdout (--out starts with '-'), the logs will be written to stderr. If the filename after --log starts with '-' it will write to stderr.

### Turn on Verbose Mode (--verbose)

Turn on verbose logging to get more log messages in the log and to stderr.

### Do not require BGZF EOF block (--noeof)

Use --noeof if you do not expect a trailing eof block in your bgzf file.

By default, the trailing empty block is expected and checked for.

### Print the Program Parameters (--params)

Use --params to print the parameters for your program to stderr.

## PhoneHome Parameters

See PhoneHome for more information on how PhoneHome works and what it does.

### Turn off PhoneHome (--noPhoneHome)

Use the --noPhoneHome option to completely disable PhoneHome. PhoneHome is enabled by default based on the thinning parameter.

### Adjust the Frequency of PhoneHome (--phoneHomeThinning)

Use --phoneHomeThinning to modify the percentage of the time that PhoneHome will run (0-100).

• By default, --phoneHomeThinning is set to 50, running 50% of the time.
• PhoneHome will only occur if the run's random number modulo 100 is less than the --phoneHomeThinning value.
• N/A if --noPhoneHome is set.

## Required Recalibration Parameters

### Reference File (--refFile)

The reference file is a required parameter used for comparing read bases to the reference.

## Optional Recalibration Parameters

### DBSNP File (--dbsnp)

The dbsnp file that specifies positions to skip recalibrating. Tab delimited file with the chromosome in the first column and the 1-based position in the 2nd column.

### Minimum Recalibration Base Quality (--minBaseQual)

When recalibrating reads, only positions with a base quality greater than this minimum phred quality will be recalibrated. If --minBaseQual is not specified, it is defaulted to 5.

The ILLUMINA specs indicate that any quality below 5 can be used as an error indicator so we do not want to recalibrate those.

### Maximum Recalibration Base Quality (--maxBaseQual)

This value sets the maximum phred base quality assigned to a base after recalibrating. Any qualities above this value will be set to this value. It is defaulted to 50.

### Blended Model Weight (--blended)

TBD - this parameter is not yet implemented.

### Fit Model (--fitModel)

Check if the logistic regression model fits the data.

This option does NOT set the new qualities to the logistic regression calculated qualities, it only checks the fit. To apply the logistic regression qualities, see --useLogReg. --fitModel is automatically applied when --useLogReg is specified.

This option cannot be used in conjunction with --fast and is overriden by --fast, but automatically applied by useLogReg

### Fast Recalibration (--fast)

Use a compact representation of the Recalibration Table that only allows:

• at most 256 Read Groups
• maximum quality 63
• at most 127 cycles

This option will run faster than the default recalibration, but uses up to about 2.25G more memory than running without --fast.

This option cannot be used in conjunction with --fitModel, or --useLogReg and overrides --fitModel, but is overridden by --useLogReg.

### Allow Previous Base DBSNP (--keepPrevDbsnp)

By default bases where the previous base is in DBSNP are excluded from the Recalibration Table.

This option includes these bases in the building of the Recalibration Table.

### Allow Previous Base Non-Match/Mismatch (--keepPrevNonAdjacent)

By default bases where the previous base is not a CIGAR Match/Mismatch are excluded from the Recalibration Table.

This option includes these bases in the building of the Recalibration Table.

### Logistic Regression (--useLogReg)

Use the logistic regression empirical qualities for setting the new base qualities instead of the default formula.

This option automatically enables --fitModel and disables --fast.

### Read the quality from a tag (--qualField)

If this parameter is set, then read the quality string from the specified tag name. If the tag is not found, the quality is read from the quality field.

### Store the original quality (--storeQualTag)

If this parameter is set, the original quality will be stored as a string in the specified tag.

### Skip Records with any of the Specified Flags (--buildExcludeFlags, --applyExcludeFlags)

Use --buildExcludeFlags to skip records with any of the specified flags set when building the recalibration table, default 0xF04.

By default, when building the recalibration table reads with any of the following flags set are skipped:

• unmapped
• secondary alignment
• fails QC checks
• duplicate
• supplementary alignment

Use --applyExcludeFlags to skip records with any of the specified flags set when applying the recalibration table. The default value is 0x000, do not skip any reads.

# Return Value

Returns -1 if input parameters are invalid.

Returns the SamStatus for the reads/writes (0 on success, non-0 on failure).