* By downloading the PROGRAM you agree to the following terms of use:

*

* BROAD INSTITUTE

* SOFTWARE LICENSE AGREEMENT

* FOR ACADEMIC NON-COMMERCIAL RESEARCH PURPOSES ONLY

*

* This Agreement is made between the Broad Institute, Inc. with a principal address at 415 Main Street, Cambridge, MA 02142 (“BROAD”) and the LICENSEE and is effective at the date the downloading is completed (“EFFECTIVE DATE”).

*

* WHEREAS, LICENSEE desires to license the PROGRAM, as defined hereinafter, and BROAD wishes to have this PROGRAM utilized in the public interest, subject only to the royalty-free, nonexclusive, nontransferable license rights of the United States Government pursuant to 48 CFR 52.227-14; and

* WHEREAS, LICENSEE desires to license the PROGRAM and BROAD desires to grant a license on the following terms and conditions.

* NOW, THEREFORE, in consideration of the promises and covenants made herein, the parties hereto agree as follows:

*

* 1. DEFINITIONS

* 1.1 PROGRAM shall mean copyright in the object code and source code known as GATK3 and related documentation, if any, as they exist on the EFFECTIVE DATE and can be downloaded from http://www.broadinstitute.org/gatk on the EFFECTIVE DATE.

*

* 2. LICENSE

* 2.1 Grant. Subject to the terms of this Agreement, BROAD hereby grants to LICENSEE, solely for academic non-commercial research purposes, a non-exclusive, non-transferable license to: (a) download, execute and display the PROGRAM and (b) create bug fixes and modify the PROGRAM. LICENSEE hereby automatically grants to BROAD a non-exclusive, royalty-free, irrevocable license to any LICENSEE bug fixes or modifications to the PROGRAM with unlimited rights to sublicense and/or distribute.  LICENSEE agrees to provide any such modifications and bug fixes to BROAD promptly upon their creation.

* The LICENSEE may apply the PROGRAM in a pipeline to data owned by users other than the LICENSEE and provide these users the results of the PROGRAM provided LICENSEE does so for academic non-commercial purposes only. For clarification purposes, academic sponsored research is not a commercial use under the terms of this Agreement.

* 2.2 No Sublicensing or Additional Rights. LICENSEE shall not sublicense or distribute the PROGRAM, in whole or in part, without prior written permission from BROAD. LICENSEE shall ensure that all of its users agree to the terms of this Agreement. LICENSEE further agrees that it shall not put the PROGRAM on a network, server, or other similar technology that may be accessed by anyone other than the LICENSEE and its employees and users who have agreed to the terms of this agreement.

* 2.3 License Limitations. Nothing in this Agreement shall be construed to confer any rights upon LICENSEE by implication, estoppel, or otherwise to any computer software, trademark, intellectual property, or patent rights of BROAD, or of any other entity, except as expressly granted herein. LICENSEE agrees that the PROGRAM, in whole or part, shall not be used for any commercial purpose, including without limitation, as the basis of a commercial software or hardware product or to provide services. LICENSEE further agrees that the PROGRAM shall not be copied or otherwise adapted in order to circumvent the need for obtaining a license for use of the PROGRAM.

*

* 3. PHONE-HOME FEATURE

* LICENSEE expressly acknowledges that the PROGRAM contains an embedded automatic reporting system (“PHONE-HOME”) which is enabled by default upon download. Unless LICENSEE requests disablement of PHONE-HOME, LICENSEE agrees that BROAD may collect limited information transmitted by PHONE-HOME regarding LICENSEE and its use of the PROGRAM.  Such information shall include LICENSEE’S user identification, version number of the PROGRAM and tools being run, mode of analysis employed, and any error reports generated during run-time.  Collection of such information is used by BROAD solely to monitor usage rates, fulfill reporting requirements to BROAD funding agencies, drive improvements to the PROGRAM, and facilitate adjustments to PROGRAM-related documentation.

*

* 4. OWNERSHIP OF INTELLECTUAL PROPERTY

* LICENSEE acknowledges that title to the PROGRAM shall remain with BROAD. The PROGRAM is marked with the following BROAD copyright notice and notice of attribution to contributors. LICENSEE shall retain such notice on all copies. LICENSEE agrees to include appropriate attribution if any results obtained from use of the PROGRAM are included in any publication.

* Copyright 2012-2014 Broad Institute, Inc.

* Notice of attribution: The GATK3 program was made available through the generosity of Medical and Population Genetics program at the Broad Institute, Inc.

* LICENSEE shall not use any trademark or trade name of BROAD, or any variation, adaptation, or abbreviation, of such marks or trade names, or any names of officers, faculty, students, employees, or agents of BROAD except as states above for attribution purposes.

*

* 5. INDEMNIFICATION

* LICENSEE shall indemnify, defend, and hold harmless BROAD, and their respective officers, faculty, students, employees, associated investigators and agents, and their respective successors, heirs and assigns, (Indemnitees), against any liability, damage, loss, or expense (including reasonable attorneys fees and expenses) incurred by or imposed upon any of the Indemnitees in connection with any claims, suits, actions, demands or judgments arising out of any theory of liability (including, without limitation, actions in the form of tort, warranty, or strict liability and regardless of whether such action has any factual basis) pursuant to any right or license granted under this Agreement.

*

* 6. NO REPRESENTATIONS OR WARRANTIES

* THE PROGRAM IS DELIVERED AS IS. BROAD MAKES NO REPRESENTATIONS OR WARRANTIES OF ANY KIND CONCERNING THE PROGRAM OR THE COPYRIGHT, EXPRESS OR IMPLIED, INCLUDING, WITHOUT LIMITATION, WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, NONINFRINGEMENT, OR THE ABSENCE OF LATENT OR OTHER DEFECTS, WHETHER OR NOT DISCOVERABLE. BROAD EXTENDS NO WARRANTIES OF ANY KIND AS TO PROGRAM CONFORMITY WITH WHATEVER USER MANUALS OR OTHER LITERATURE MAY BE ISSUED FROM TIME TO TIME.

* IN NO EVENT SHALL BROAD OR ITS RESPECTIVE DIRECTORS, OFFICERS, EMPLOYEES, AFFILIATED INVESTIGATORS AND AFFILIATES BE LIABLE FOR INCIDENTAL OR CONSEQUENTIAL DAMAGES OF ANY KIND, INCLUDING, WITHOUT LIMITATION, ECONOMIC DAMAGES OR INJURY TO PROPERTY AND LOST PROFITS, REGARDLESS OF WHETHER BROAD SHALL BE ADVISED, SHALL HAVE OTHER REASON TO KNOW, OR IN FACT SHALL KNOW OF THE POSSIBILITY OF THE FOREGOING.

*

* 7. ASSIGNMENT

* This Agreement is personal to LICENSEE and any rights or obligations assigned by LICENSEE without the prior written consent of BROAD shall be null and void.

*

* 8. MISCELLANEOUS

* 8.1 Export Control. LICENSEE gives assurance that it will comply with all United States export control laws and regulations controlling the export of the PROGRAM, including, without limitation, all Export Administration Regulations of the United States Department of Commerce. Among other things, these laws and regulations prohibit, or require a license for, the export of certain types of software to specified countries.

* 8.2 Termination. LICENSEE shall have the right to terminate this Agreement for any reason upon prior written notice to BROAD. If LICENSEE breaches any provision hereunder, and fails to cure such breach within thirty (30) days, BROAD may terminate this Agreement immediately. Upon termination, LICENSEE shall provide BROAD with written assurance that the original and all copies of the PROGRAM have been destroyed, except that, upon prior written authorization from BROAD, LICENSEE may retain a copy for archive purposes.

* 8.3 Survival. The following provisions shall survive the expiration or termination of this Agreement: Articles 1, 3, 4, 5 and Sections 2.2, 2.3, 7.3, and 7.4.

* 8.4 Notice. Any notices under this Agreement shall be in writing, shall specifically refer to this Agreement, and shall be sent by hand, recognized national overnight courier, confirmed facsimile transmission, confirmed electronic mail, or registered or certified mail, postage prepaid, return receipt requested. All notices under this Agreement shall be deemed effective upon receipt.

* 8.5 Amendment and Waiver; Entire Agreement. This Agreement may be amended, supplemented, or otherwise modified only by means of a written instrument signed by all parties. Any waiver of any rights or failure to act in a specific instance shall relate only to such instance and shall not be construed as an agreement to waive any rights or fail to act in any other instance, whether or not similar. This Agreement constitutes the entire agreement among the parties with respect to its subject matter and supersedes prior agreements or understandings between the parties relating to its subject matter.

* 8.6 Binding Effect; Headings. This Agreement shall be binding upon and inure to the benefit of the parties and their respective permitted successors and assigns. All headings are for convenience only and shall not affect the meaning of any provision of this Agreement.

* 8.7 Governing Law. This Agreement shall be construed, governed, interpreted and applied in accordance with the internal laws of the Commonwealth of Massachusetts, U.S.A., without regard to conflict of laws principles.

*/

* Performs local realignment of reads to correct misalignments due to the presence of indels.

*

* <p>

* The local realignment tool is designed to consume one or more BAM files and to locally realign reads such that the number of mismatching bases

* is minimized across all the reads. In general, a large percent of regions requiring local realignment are due to the presence of an insertion

* or deletion (indels) in the individual's genome with respect to the reference genome.  Such alignment artifacts result in many bases mismatching

* the reference near the misalignment, which are easily mistaken as SNPs.  Moreover, since read mapping algorithms operate on each read independently,

* it is impossible to place reads on the reference genome such at mismatches are minimized across all reads.  Consequently, even when some reads are

* correctly mapped with indels, reads covering the indel near just the start or end of the read are often incorrectly mapped with respect the true indel,

* also requiring realignment.  Local realignment serves to transform regions with misalignments due to indels into clean reads containing a consensus

* indel suitable for standard variant discovery approaches.  Unlike most mappers, this walker uses the full alignment context to determine whether an

* appropriate alternate reference (i.e. indel) exists.  Following local realignment, the GATK tool Unified Genotyper can be used to sensitively and

* specifically identify indels.

* </p>

*     <ol>There are 2 steps to the realignment process:

*     <li>Determining (small) suspicious intervals which are likely in need of realignment (see the RealignerTargetCreator tool)</li>

*     <li>Running the realigner over those intervals (IndelRealigner)</li>

*     </ol>

* <p>

* For more details, see http://www.broadinstitute.org/gatk/guide/article?id=38

* </p>

*

* <h3>Input</h3>

* <p>

* One or more aligned BAM files and optionally one or more lists of known indels.

* </p>

*

* <h3>Output</h3>

* <p>

* A realigned version of your input BAM file(s).

* </p>

*

* <h3>Example</h3>

* <pre>

* java -Xmx4g -jar GenomeAnalysisTK.jar \

*   -T IndelRealigner \

*   -R ref.fasta \

*   -I input.bam \

*   -targetIntervals intervalListFromRTC.intervals \

*   -o realignedBam.bam \

*   [-known /path/to/indels.vcf] \

*   [-compress 0]    (this argument recommended to speed up the process *if* this is only a temporary file; otherwise, use the default value)

* </pre>

*

* <h3>Caveats</h3>

*

* <ul><li>

* An important note: the input bam(s), reference, and known indel file(s) should be the same ones used for the RealignerTargetCreator step.

* </li><li>

* Another important note: because reads produced from the 454 technology inherently contain false indels, the realigner will not currently work with them

* (or with reads from similar technologies).

* </li></ul>

*

* @author ebanks

*/

    public static final String ORIGINAL_CIGAR_TAG = "OC";

    public static final String ORIGINAL_POSITION_TAG = "OP";

    public static final String PROGRAM_RECORD_NAME = "GATK IndelRealigner";

    public enum ConsensusDeterminationModel {

        /**

         * Uses only indels from a provided ROD of known indels.

         */

        KNOWNS_ONLY,

        /**

         * Additionally uses indels already present in the original alignments of the reads.

         */

        USE_READS,

        /**

         * Additionally uses 'Smith-Waterman' to generate alternate consenses.

         */

        USE_SW

    }

    /**

     * Any number of VCF files representing known indels to be used for constructing alternate consenses.

     * Could be e.g. dbSNP and/or official 1000 Genomes indel calls.  Non-indel variants in these files will be ignored.

     */

    @Input(fullName="knownAlleles", shortName = "known", doc="Input VCF file(s) with known indels", required=false)

    public List<RodBinding<VariantContext>> known = Collections.emptyList();

    /**

     * The interval list output from the RealignerTargetCreator tool using the same bam(s), reference, and known indel file(s).

     */

    @Input(fullName="targetIntervals", shortName="targetIntervals", doc="Intervals file output from RealignerTargetCreator", required=true)

    protected IntervalBinding<Feature> intervalsFile = null;

    /**

     * This term is equivalent to "significance" - i.e. is the improvement significant enough to merit realignment? Note that this number

     * should be adjusted based on your particular data set. For low coverage and/or when looking for indels with low allele frequency,

     * this number should be smaller.

     */

    @Argument(fullName="LODThresholdForCleaning", shortName="LOD", doc="LOD threshold above which the cleaner will clean", required=false)

    protected double LOD_THRESHOLD = 5.0;

    /**

     * The realigned bam file.

     */

    @Output(required=false, doc="Output bam", defaultToStdout=false)

    protected GATKSAMFileWriter writer = null;

    protected ConstrainedMateFixingManager manager = null;

    protected SAMFileWriter writerToUse = null;

    /**

     * We recommend that users run with USE_READS when trying to realign high quality longer read data mapped with a gapped aligner;

     * Smith-Waterman is really only necessary when using an ungapped aligner (e.g. MAQ in the case of single-end read data).

     */

    @Argument(fullName = "consensusDeterminationModel", shortName = "model", doc = "Determines how to compute the possible alternate consenses", required = false)

    public ConsensusDeterminationModel consensusModel = ConsensusDeterminationModel.USE_READS;

    // ADVANCED OPTIONS FOLLOW

    /**

     * For expert users only!  This is similar to the argument in the RealignerTargetCreator walker. The point here is that the realigner

     * will only proceed with the realignment (even above the given threshold) if it minimizes entropy among the reads (and doesn't simply

     * push the mismatch column to another position). This parameter is just a heuristic and should be adjusted based on your particular data set.

     */

    @Advanced

    @Argument(fullName="entropyThreshold", shortName="entropy", doc="Percentage of mismatches at a locus to be considered having high entropy (0.0 < entropy <= 1.0)", required=false)

    protected double MISMATCH_THRESHOLD = 0.15;

    /**

     * For expert users only!  To minimize memory consumption you can lower this number (but then the tool may skip realignment on regions with too much coverage;

     * and if the number is too low, it may generate errors during realignment). Just make sure to give Java enough memory! 4Gb should be enough with the default value.

     */

    @Advanced

    @Argument(fullName="maxReadsInMemory", shortName="maxInMemory", doc="max reads allowed to be kept in memory at a time by the SAMFileWriter", required=false)

    protected int MAX_RECORDS_IN_MEMORY = 150000;

    /**

     * For expert users only!

     */

    @Advanced

    @Argument(fullName="maxIsizeForMovement", shortName="maxIsize", doc="maximum insert size of read pairs that we attempt to realign", required=false)

    protected int MAX_ISIZE_FOR_MOVEMENT = 3000;

    /**

     * For expert users only!

     */

    @Advanced

    @Argument(fullName="maxPositionalMoveAllowed", shortName="maxPosMove", doc="Maximum positional move in basepairs that a read can be adjusted during realignment", required=false)

    protected int MAX_POS_MOVE_ALLOWED = 200;

    /**

     * For expert users only!  If you need to find the optimal solution regardless of running time, use a higher number.

     */

    @Advanced

    @Argument(fullName="maxConsensuses", shortName="maxConsensuses", doc="Max alternate consensuses to try (necessary to improve performance in deep coverage)", required=false)

    protected int MAX_CONSENSUSES = 30;

    /**

     * For expert users only!  If you need to find the optimal solution regardless of running time, use a higher number.

     */

    @Advanced

    @Argument(fullName="maxReadsForConsensuses", shortName="greedy", doc="Max reads used for finding the alternate consensuses (necessary to improve performance in deep coverage)", required=false)

    protected int MAX_READS_FOR_CONSENSUSES = 120;

    /**

     * For expert users only!  If this value is exceeded at a given interval, realignment is not attempted and the reads are passed to the output file(s) as-is.

     * If you need to allow more reads (e.g. with very deep coverage) regardless of memory, use a higher number.

     */

    @Advanced

    @Argument(fullName="maxReadsForRealignment", shortName="maxReads", doc="Max reads allowed at an interval for realignment", required=false)

    protected int MAX_READS = 20000;

    @Advanced

    @Argument(fullName="noOriginalAlignmentTags", shortName="noTags", required=false, doc="Don't output the original cigar or alignment start tags for each realigned read in the output bam")

    protected boolean NO_ORIGINAL_ALIGNMENT_TAGS = false;

    /**

     * Reads from all input files will be realigned together, but then each read will be saved in the output file corresponding to the input file that

     * the read came from. There are two ways to generate output bam file names: 1) if the value of this argument is a general string (e.g. '.cleaned.bam'),

     * then extensions (".bam" or ".sam") will be stripped from the input file names and the provided string value will be pasted on instead; 2) if the

     * value ends with a '.map' (e.g. input_output.map), then the two-column tab-separated file with the specified name must exist and list unique output

     * file name (2nd column) for each input file name (1st column).

     *

     * Note that some GATK arguments do NOT work in conjunction with nWayOut (e.g. --disable_bam_indexing).

     */

    @Argument(fullName="nWayOut", shortName="nWayOut", required=false, doc="Generate one output file for each input (-I) bam file (not compatible with -output)")

    protected String N_WAY_OUT = null;

    @Hidden

    @Argument(fullName="generate_nWayOut_md5s",doc="Generate md5sums for BAMs")

    protected boolean generateMD5s = false;

    // DEBUGGING OPTIONS FOLLOW

    @Hidden

    @Argument(fullName="check_early",shortName="check_early",required=false,doc="Do early check of reads against existing consensuses")

    protected boolean CHECKEARLY = false;

    @Hidden

    @Argument(fullName="noPGTag", shortName="noPG", required=false,

            doc="Don't output the usual PG tag in the realigned bam file header. FOR DEBUGGING PURPOSES ONLY.  This option is required in order to pass integration tests.")

    protected boolean NO_PG_TAG = false;

    @Hidden

    @Argument(fullName="keepPGTags", shortName="keepPG", required=false,

            doc="Keep older PG tags left in the bam header by previous runs of this tool (by default, all these "+

                    "historical tags will be replaced by the latest tag generated in the current run).")

    protected boolean KEEP_ALL_PG_RECORDS = false;

    @Hidden

    @Output(fullName="indelsFileForDebugging", shortName="indels", required=false, defaultToStdout=false, doc="Output file (text) for the indels found; FOR DEBUGGING PURPOSES ONLY")

    protected String OUT_INDELS = null;

    @Hidden

    @Output(fullName="statisticsFileForDebugging", shortName="stats", doc="print out statistics (what does or doesn't get cleaned); FOR DEBUGGING PURPOSES ONLY", required=false, defaultToStdout=false)

    protected String OUT_STATS = null;

    @Hidden

    @Output(fullName="SNPsFileForDebugging", shortName="snps", doc="print out whether mismatching columns do or don't get cleaned out; FOR DEBUGGING PURPOSES ONLY", required=false, defaultToStdout=false)

    protected String OUT_SNPS = null;

    // fasta reference reader to supplement the edges of the reference sequence

    private CachingIndexedFastaSequenceFile referenceReader;

    // the intervals input by the user

    private Iterator<GenomeLoc> intervals = null;

    // the current interval in the list

    private GenomeLoc currentInterval = null;

    private boolean sawReadInCurrentInterval = false;

    // the reads and known indels that fall into the current interval

    private ReadBin readsToClean;

    private final ArrayList<GATKSAMRecord> readsNotToClean = new ArrayList<GATKSAMRecord>();

    private final ArrayList<VariantContext> knownIndelsToTry = new ArrayList<VariantContext>();

    private final HashSet<Object> indelRodsSeen = new HashSet<Object>();

    private final HashSet<GATKSAMRecord> readsActuallyCleaned = new HashSet<GATKSAMRecord>();

    private static final int MAX_QUAL = 99;

    // fraction of mismatches that need to no longer mismatch for a column to be considered cleaned

    private static final double MISMATCH_COLUMN_CLEANED_FRACTION = 0.75;

    private final static Parameters swParameters = new Parameters(30, -10, -10, -2);

    // reference base padding size

    // TODO -- make this a command-line argument if the need arises

    private static final int REFERENCE_PADDING = 30;

    // other output files

    private FileWriter indelOutput = null;

    private FileWriter statsOutput = null;

    private FileWriter snpsOutput = null;

    //###protected Map<SAMReaderID, ConstrainedMateFixingManager> nwayWriters = null;

    // debug info for lazy SW evaluation:

    private long exactMatchesFound = 0; // how many reads exactly matched a consensus we already had

    private long SWalignmentRuns = 0; // how many times (=for how many reads) we ran SW alignment

    private long SWalignmentSuccess = 0; // how many SW alignments were "successful" (i.e. found a workable indel and resulted in non-null consensus)

    private Map<String,String> loadFileNameMap(String mapFile) {

        Map<String,String> fname_map = new HashMap<String,String>();

        try {

             XReadLines reader = new XReadLines(new File(mapFile),true);

             for ( String line : reader ) {

                 if ( line.length() == 0 ) continue;

                 String fields[] = line.split("\t");

                 if ( fields.length != 2 )

                     throw new UserException.BadInput("Input-output map file must have exactly two columns. Offending line:\n"+line);

                 if ( fields[0].length() == 0 || fields[1].length() == 0 )

                     throw new UserException.BadInput("Input-output map file can not have empty strings in either column. Offending line:\n"+line);

                 if ( fname_map.containsKey(fields[0]) )

                     throw new UserException.BadInput("Input-output map file contains duplicate entries for input name "+fields[0]);

                 if ( fname_map.containsValue(fields[1]) )

                     throw new UserException.BadInput("Input-output map file maps multiple entries onto single output name "+fields[1]);

                 fname_map.put(fields[0],fields[1]);

             }

        } catch (IOException e) {

            throw new GATKException("I/O Error while reading input-output map file "+N_WAY_OUT+": "+e.getMessage());

        }

       return fname_map;

    }

    public void initialize() {

        readsToClean = new ReadBin(getToolkit().getGenomeLocParser(), REFERENCE_PADDING);

        if ( N_WAY_OUT == null && writer == null ) {

            throw new UserException.CommandLineException("Either -o or -nWayOut must be specified");

        }

        if ( N_WAY_OUT != null && writer != null ) {

            throw new UserException.CommandLineException("-o and -nWayOut can not be used simultaneously");

        }

        if ( LOD_THRESHOLD < 0.0 )

            throw new RuntimeException("LOD threshold cannot be a negative number");

        if ( MISMATCH_THRESHOLD <= 0.0 || MISMATCH_THRESHOLD > 1.0 )

            throw new RuntimeException("Entropy threshold must be a fraction between 0 and 1");

        try {

            referenceReader = new CachingIndexedFastaSequenceFile(getToolkit().getArguments().referenceFile);

        }

        catch(FileNotFoundException ex) {

            throw new UserException.CouldNotReadInputFile(getToolkit().getArguments().referenceFile,ex);

        }

        intervals = intervalsFile.getIntervals(getToolkit()).iterator();

        currentInterval = intervals.hasNext() ? intervals.next() : null;

        if ( N_WAY_OUT != null ) {

            boolean createIndex =  true;

            if ( N_WAY_OUT.toUpperCase().endsWith(".MAP") ) {

                writerToUse = new NWaySAMFileWriter(getToolkit(),loadFileNameMap(N_WAY_OUT),

                            SAMFileHeader.SortOrder.coordinate,true, createIndex, generateMD5s,createProgramRecord(),KEEP_ALL_PG_RECORDS);

            } else {

                writerToUse = new NWaySAMFileWriter(getToolkit(),N_WAY_OUT,SAMFileHeader.SortOrder.coordinate,true,

                        createIndex, generateMD5s,createProgramRecord(),KEEP_ALL_PG_RECORDS);

            }

        }   else {

            // set up the output writer

            setupWriter(getToolkit().getSAMFileHeader());

            writerToUse = writer;

        }

        manager = new ConstrainedMateFixingManager(writerToUse, getToolkit().getGenomeLocParser(), MAX_ISIZE_FOR_MOVEMENT, MAX_POS_MOVE_ALLOWED, MAX_RECORDS_IN_MEMORY);

        if ( OUT_INDELS != null ) {

            try {

                indelOutput = new FileWriter(new File(OUT_INDELS));

            } catch (Exception e) {

                logger.error("Failed to create output file "+ OUT_INDELS+". Indel output will be suppressed");

                logger.error(e.getMessage());

                indelOutput = null;

            }

        }

        if ( OUT_STATS != null ) {

            try {

                statsOutput = new FileWriter(new File(OUT_STATS));

            } catch (Exception e) {

                logger.error("Failed to create output file "+ OUT_STATS+". Cleaning stats output will be suppressed");

                logger.error(e.getMessage());

                statsOutput = null;

            }

        }

        if ( OUT_SNPS != null ) {

            try {

                snpsOutput = new FileWriter(new File(OUT_SNPS));

            } catch (Exception e) {

                logger.error("Failed to create output file "+ OUT_SNPS+". Cleaning snps output will be suppressed");

                logger.error(e.getMessage());

                snpsOutput = null;

            }

        }

    }

    private void setupWriter(SAMFileHeader header) {

       

        if ( !NO_PG_TAG ) {

            final SAMProgramRecord programRecord = createProgramRecord();

            List<SAMProgramRecord> oldRecords = header.getProgramRecords();

            List<SAMProgramRecord> newRecords = new ArrayList<SAMProgramRecord>(oldRecords.size()+1);

            for ( SAMProgramRecord record : oldRecords ) {

                if ( !record.getId().startsWith(PROGRAM_RECORD_NAME) || KEEP_ALL_PG_RECORDS )

                    newRecords.add(record);

            }

            newRecords.add(programRecord);

            header.setProgramRecords(newRecords);

        }

        writer.writeHeader(header);

        writer.setPresorted(true);

    }

    private SAMProgramRecord createProgramRecord() {

        if ( NO_PG_TAG ) return null;

        final SAMProgramRecord programRecord = new SAMProgramRecord(PROGRAM_RECORD_NAME);

        final ResourceBundle headerInfo = TextFormattingUtils.loadResourceBundle("StingText");

        try {

            final String version = headerInfo.getString("org.broadinstitute.sting.gatk.version");

            programRecord.setProgramVersion(version);

        } catch (MissingResourceException e) {

            // this is left empty on purpose (perhaps Andrey knows why?)

        }

        programRecord.setCommandLine(getToolkit().createApproximateCommandLineArgumentString(getToolkit(), this));

        return programRecord;

    }

    private void emit(final GATKSAMRecord read) {

        // check to see whether the read was modified by looking at the temporary tag

        boolean wasModified = readsActuallyCleaned.contains(read);

        try {

            manager.addRead(read, wasModified);

        } catch (RuntimeIOException e) {

            throw new UserException.ErrorWritingBamFile(e.getMessage());

        }

    }

    private void emitReadLists() {

        // pre-merge lists to sort them in preparation for constrained SAMFileWriter

        readsNotToClean.addAll(readsToClean.getReads());

        ReadUtils.sortReadsByCoordinate(readsNotToClean);

        manager.addReads(readsNotToClean, readsActuallyCleaned);

        readsToClean.clear();

        readsNotToClean.clear();

        readsActuallyCleaned.clear();

    }

    public Integer map(ReferenceContext ref, GATKSAMRecord read, RefMetaDataTracker metaDataTracker) {

        if ( currentInterval == null ) {

            emit(read);

            return 0;

        }

        // edge case: when the last target interval abuts the end of the genome, we'll get one of the

        //   unmapped reads while the currentInterval still isn't null.  We need to trigger the cleaning

        //   at this point without trying to create a GenomeLoc.

        if ( read.getReferenceIndex() == SAMRecord.NO_ALIGNMENT_REFERENCE_INDEX ) {

            cleanAndCallMap(ref, read, metaDataTracker, null);

            return 0;

        }

        GenomeLoc readLoc = getToolkit().getGenomeLocParser().createGenomeLoc(read);

        // hack to get around unmapped reads having screwy locations

        if ( readLoc.getStop() == 0 )

            readLoc = getToolkit().getGenomeLocParser().createGenomeLoc(readLoc.getContig(), readLoc.getStart(), readLoc.getStart());

        if ( readLoc.isBefore(currentInterval) ) {

            if ( !sawReadInCurrentInterval )

                emit(read);

            else

                readsNotToClean.add(read);

        }

        else if ( readLoc.overlapsP(currentInterval) ) {

            sawReadInCurrentInterval = true;

            if ( doNotTryToClean(read) ) {

                readsNotToClean.add(read);

            } else {

                readsToClean.add(read);

                // add the rods to the list of known variants

                populateKnownIndels(metaDataTracker);

            }

            if ( readsToClean.size() + readsNotToClean.size() >= MAX_READS ) {

                logger.info("Not attempting realignment in interval " + currentInterval + " because there are too many reads.");

                abortCleanForCurrentInterval();

            }

        }

        else {  // the read is past the current interval

            logger.debug(currentInterval.toString() + "\t" + read.getAlignmentStart() );

            cleanAndCallMap(ref, read, metaDataTracker, readLoc);

        }

        return 0;

    }

    private void abortCleanForCurrentInterval() {

        emitReadLists();

        currentInterval = intervals.hasNext() ? intervals.next() : null;

        sawReadInCurrentInterval = false;

    }

    private boolean doNotTryToClean(GATKSAMRecord read) {

        return read.getReadUnmappedFlag() ||

                read.getNotPrimaryAlignmentFlag() ||

                read.getReadFailsVendorQualityCheckFlag() ||

                read.getMappingQuality() == 0 ||

                read.getAlignmentStart() == SAMRecord.NO_ALIGNMENT_START ||

                ConstrainedMateFixingManager.iSizeTooBigToMove(read, MAX_ISIZE_FOR_MOVEMENT) ||

                ReadUtils.is454Read(read) ||

                ReadUtils.isIonRead(read);

        // TODO -- it would be nice if we could use indels from 454/Ion reads as alternate consenses

    }

    private void cleanAndCallMap(ReferenceContext ref, GATKSAMRecord read, RefMetaDataTracker metaDataTracker, GenomeLoc readLoc) {

        if ( readsToClean.size() > 0 ) {

            GenomeLoc earliestPossibleMove = getToolkit().getGenomeLocParser().createGenomeLoc(readsToClean.getReads().get(0));

            if ( manager.canMoveReads(earliestPossibleMove) )

                clean(readsToClean);

        }

        knownIndelsToTry.clear();

        indelRodsSeen.clear();

        emitReadLists();

        try {

            do {

                currentInterval = intervals.hasNext() ? intervals.next() : null;

            } while ( currentInterval != null && (readLoc == null || currentInterval.isBefore(readLoc)) );

        } catch (ReviewedGATKException e) {

            throw new UserException.MissortedFile(new File(intervalsFile.getSource()), " *** Are you sure that your interval file is sorted? If not, you must use the --targetIntervalsAreNotSorted argument. ***", e);

        }

        sawReadInCurrentInterval = false;

        // call back into map now that the state has been updated

        map(ref, read, metaDataTracker);

    }

    public Integer reduceInit() {

        return 0;

    }

    public Integer reduce(Integer value, Integer sum) {

        return sum + value;

    }

    public void onTraversalDone(Integer result) {

        if ( readsToClean.size() > 0 ) {

            GenomeLoc earliestPossibleMove = getToolkit().getGenomeLocParser().createGenomeLoc(readsToClean.getReads().get(0));

            if ( manager.canMoveReads(earliestPossibleMove) )

                clean(readsToClean);

            emitReadLists();

        } else if ( readsNotToClean.size() > 0 ) {

            emitReadLists();                           

        }

        knownIndelsToTry.clear();

        indelRodsSeen.clear();

        if ( OUT_INDELS != null ) {

            try {

                indelOutput.close();

            } catch (Exception e) {

                logger.error("Failed to close "+OUT_INDELS+" gracefully. Data may be corrupt.");

            }

        }

        if ( OUT_STATS != null ) {

            try {

                statsOutput.close();

            } catch (Exception e) {

                logger.error("Failed to close "+OUT_STATS+" gracefully. Data may be corrupt.");

            }

        }

        if ( OUT_SNPS != null ) {

            try {

                snpsOutput.close();

            } catch (Exception e) {

                logger.error("Failed to close "+OUT_SNPS+" gracefully. Data may be corrupt.");

            }

        }

        manager.close();

        if ( N_WAY_OUT != null ) writerToUse.close();

        if ( CHECKEARLY ) {

            logger.info("SW alignments runs: "+SWalignmentRuns);

            logger.info("SW alignments successfull: "+SWalignmentSuccess + " ("+SWalignmentSuccess/SWalignmentRuns+"% of SW runs)");

            logger.info("SW alignments skipped (perfect match): "+exactMatchesFound);

            logger.info("Total reads SW worked for: "+(SWalignmentSuccess + exactMatchesFound)+

                    " ("+(SWalignmentSuccess+exactMatchesFound)/(SWalignmentRuns+exactMatchesFound)+"% of all reads requiring SW)");

        }

    }

    private void populateKnownIndels(RefMetaDataTracker metaDataTracker) {

        for ( final VariantContext vc : metaDataTracker.getValues(known) ) {

            if ( indelRodsSeen.contains(vc) )

                continue;

            indelRodsSeen.add(vc);

            knownIndelsToTry.add(vc);

        }

    }

    private static int mismatchQualitySumIgnoreCigar(final AlignedRead aRead, final byte[] refSeq, int refIndex, int quitAboveThisValue) {

        final byte[] readSeq = aRead.getReadBases();

        final byte[] quals = aRead.getBaseQualities();

        int sum = 0;

        for (int readIndex = 0 ; readIndex < readSeq.length ; refIndex++, readIndex++ ) {

            if ( refIndex >= refSeq.length ) {

                sum += MAX_QUAL;

                // optimization: once we pass the threshold, stop calculating

                if ( sum > quitAboveThisValue )

                    return sum;

            } else {

                byte refChr = refSeq[refIndex];

                byte readChr = readSeq[readIndex];

                if ( !BaseUtils.isRegularBase(readChr) || !BaseUtils.isRegularBase(refChr) )

                    continue; // do not count Ns/Xs/etc ?

                if ( readChr != refChr ) {

                    sum += (int)quals[readIndex];

                    // optimization: once we pass the threshold, stop calculating

                    if ( sum > quitAboveThisValue )

                        return sum;

                }

            }

        }

        return sum;

    }

    private void clean(ReadBin readsToClean) {

        final List<GATKSAMRecord> reads = readsToClean.getReads();

        if ( reads.size() == 0 )

            return;

        byte[] reference = readsToClean.getReference(referenceReader);

        int leftmostIndex = readsToClean.getLocation().getStart();

        final ArrayList<GATKSAMRecord> refReads = new ArrayList<GATKSAMRecord>();                 // reads that perfectly match ref

        final ArrayList<AlignedRead> altReads = new ArrayList<AlignedRead>();               // reads that don't perfectly match

        final LinkedList<AlignedRead> altAlignmentsToTest = new LinkedList<AlignedRead>();  // should we try to make an alt consensus from the read?

        final Set<Consensus> altConsenses = new LinkedHashSet<Consensus>();               // list of alt consenses

        // if there are any known indels for this region, get them and create alternate consenses

        generateAlternateConsensesFromKnownIndels(altConsenses, leftmostIndex, reference);

        // decide which reads potentially need to be cleaned;

        // if there are reads with a single indel in them, add that indel to the list of alternate consenses

        long totalRawMismatchSum = determineReadsThatNeedCleaning(reads, refReads, altReads, altAlignmentsToTest, altConsenses, leftmostIndex, reference);

        // use 'Smith-Waterman' to create alternate consenses from reads that mismatch the reference, using totalRawMismatchSum as the random seed

        if ( consensusModel == ConsensusDeterminationModel.USE_SW )

            generateAlternateConsensesFromReads(altAlignmentsToTest, altConsenses, reference, leftmostIndex);

        // if ( debugOn ) System.out.println("------\nChecking consenses...\n--------\n");

        Consensus bestConsensus = null;

        for (Consensus consensus : altConsenses) {

            //logger.debug("Trying new consensus: " + consensus.cigar + " " + new String(consensus.str));

//            if ( DEBUG ) {

//                System.out.println("Checking consensus with alignment at "+consensus.positionOnReference+" cigar "+consensus.cigar);

//                System.out.println(new String(consensus.str));

//                int z = 0;

//                for ( ; z < consensus.positionOnReference; z++ )  System.out.print('.');

//                for ( z=0 ; z < consensus.cigar.getCigarElement(0).getLength() ; z++ ) System.out.print('.');

//                if ( consensus.cigar.getCigarElement(1).getOperator() == CigarOperator.I ) for ( z= 0; z < consensus.cigar.getCigarElement(1).getLength(); z++ )  System.out.print('I');

//                System.out.println();

//            }

            // if ( debugOn ) System.out.println("Consensus: "+consensus.str);

            for (int j = 0; j < altReads.size(); j++) {

                AlignedRead toTest = altReads.get(j);

                Pair<Integer, Integer> altAlignment = findBestOffset(consensus.str, toTest, leftmostIndex);

                // the mismatch score is the min of its alignment vs. the reference and vs. the alternate

                int myScore = altAlignment.second;

                if (myScore > toTest.getAlignerMismatchScore() || myScore >= toTest.getMismatchScoreToReference())

                    myScore = toTest.getMismatchScoreToReference();

                    // keep track of reads that align better to the alternate consensus.

                    // By pushing alignments with equal scores to the alternate, it means we'll over-call (het -> hom non ref) but are less likely to under-call (het -> ref, het non ref -> het)

                else

                    consensus.readIndexes.add(new Pair<Integer, Integer>(j, altAlignment.first));

                //logger.debug(consensus.cigar +  " vs. " + toTest.getRead().getReadName() + "-" + toTest.getRead().getReadString() + " => " + myScore + " vs. " + toTest.getMismatchScoreToReference());

                if (!toTest.getRead().getDuplicateReadFlag())

                    consensus.mismatchSum += myScore;

                // optimization: once the mismatch sum is higher than the best consensus, quit since this one can't win

                //  THIS MUST BE DISABLED IF WE DECIDE TO ALLOW MORE THAN ONE ALTERNATE CONSENSUS!

                if (bestConsensus != null && consensus.mismatchSum > bestConsensus.mismatchSum)

                    break;

            }

            //logger.debug("Mismatch sum of new consensus: " + consensus.mismatchSum);

            if (bestConsensus == null || bestConsensus.mismatchSum > consensus.mismatchSum) {

                // we do not need this alt consensus, release memory right away!!

                if (bestConsensus != null)

                    bestConsensus.readIndexes.clear();

                bestConsensus = consensus;

                //logger.debug("New consensus " + bestConsensus.cigar +  " is now best consensus");

            } else {

                // we do not need this alt consensus, release memory right away!!

                consensus.readIndexes.clear();

            }

        }

        // if:

        // 1) the best alternate consensus has a smaller sum of quality score mismatches than the aligned version of the reads,

        // 2) beats the LOD threshold for the sum of quality score mismatches of the raw version of the reads,

        // 3) didn't just move around the mismatching columns (i.e. it actually reduces entropy),

        // then clean!

        final double improvement = (bestConsensus == null ? -1 : ((double)(totalRawMismatchSum - bestConsensus.mismatchSum))/10.0);

        if ( improvement >= LOD_THRESHOLD ) {

            bestConsensus.cigar = AlignmentUtils.leftAlignIndel(bestConsensus.cigar, reference, bestConsensus.str, bestConsensus.positionOnReference, bestConsensus.positionOnReference, true);

           // start cleaning the appropriate reads

            for ( Pair<Integer, Integer> indexPair : bestConsensus.readIndexes ) {

                AlignedRead aRead = altReads.get(indexPair.first);

                if ( !updateRead(bestConsensus.cigar, bestConsensus.positionOnReference, indexPair.second, aRead, leftmostIndex) )

                    return;

            }

            if ( consensusModel != ConsensusDeterminationModel.KNOWNS_ONLY && !alternateReducesEntropy(altReads, reference, leftmostIndex) ) {

                if ( statsOutput != null ) {

                    try {

                        statsOutput.write(currentInterval.toString());

                        statsOutput.write("\tFAIL (bad indel)\t"); // if improvement > LOD_THRESHOLD *BUT* entropy is not reduced (SNPs still exist)

                        statsOutput.write(Double.toString(improvement));

                        statsOutput.write("\n");

                        statsOutput.flush();

                    } catch (Exception e) {

                        throw new UserException.CouldNotCreateOutputFile("statsOutput", "Failed to write stats output file", e);

                    }

                }

            } else {

                //logger.debug("CLEAN: " + bestConsensus.cigar + " " + bestConsensus.str.toString() + " " + bestConsensus.cigar.numCigarElements() );

                if ( indelOutput != null && bestConsensus.cigar.numCigarElements() > 1 ) {

                    // NOTE: indels are printed out in the format specified for the low-coverage pilot1

                    //  indel calls (tab-delimited): chr position size type sequence

                    StringBuilder str = new StringBuilder();

                    str.append(reads.get(0).getReferenceName());

                    int position = bestConsensus.positionOnReference + bestConsensus.cigar.getCigarElement(0).getLength();

                    str.append("\t").append(leftmostIndex + position - 1);

                    CigarElement ce = bestConsensus.cigar.getCigarElement(1);

                    str.append("\t").append(ce.getLength()).append("\t").append(ce.getOperator()).append("\t");

                    int length = ce.getLength();

                    if ( ce.getOperator() == CigarOperator.D ) {

                        for ( int i = 0; i < length; i++)

                            str.append((char)reference[position+i]);

                    } else {

                        for ( int i = 0; i < length; i++)

                            str.append((char)bestConsensus.str[position+i]);

                    }

                    str.append("\t").append((((double) (totalRawMismatchSum - bestConsensus.mismatchSum)) / 10.0)).append("\n");

                    try {

                        indelOutput.write(str.toString());

                        indelOutput.flush();

                    } catch (Exception e) {

                        throw new UserException.CouldNotCreateOutputFile("indelOutput", "Failed to write indel output file", e);

                    }

                }

                if ( statsOutput != null ) {

                    try {

                        statsOutput.write(currentInterval.toString());

                        statsOutput.write("\tCLEAN"); // if improvement > LOD_THRESHOLD *AND* entropy is reduced

                        if ( bestConsensus.cigar.numCigarElements() > 1 )

                            statsOutput.write(" (found indel)");

                        statsOutput.write("\t");

                        statsOutput.write(Double.toString(improvement));

                        statsOutput.write("\n");

                        statsOutput.flush();

                    } catch (Exception e) {

                        throw new UserException.CouldNotCreateOutputFile("statsOutput", "Failed to write stats output file", e);

                    }

                }

                // finish cleaning the appropriate reads

                for ( Pair<Integer, Integer> indexPair : bestConsensus.readIndexes ) {

                    final AlignedRead aRead = altReads.get(indexPair.first);

                    if ( aRead.finalizeUpdate() ) {

                        // We need to update the mapping quality score of the cleaned reads;

                        // however we don't have enough info to use the proper MAQ scoring system.

                        // For now, we will just arbitrarily add 10 to the mapping quality. [EB, 6/7/2010].

                        // TODO -- we need a better solution here

                        GATKSAMRecord read = aRead.getRead();

                        if ( read.getMappingQuality() != 255 ) // 255 == Unknown, so don't modify it

                            read.setMappingQuality(Math.min(aRead.getRead().getMappingQuality() + 10, 254));

                        // before we fix the attribute tags we first need to make sure we have enough of the reference sequence

                        int neededBasesToLeft = leftmostIndex - read.getAlignmentStart();

                        int neededBasesToRight = read.getAlignmentEnd() - leftmostIndex - reference.length + 1;

                        int neededBases = Math.max(neededBasesToLeft, neededBasesToRight);

                        if ( neededBases > 0 ) {

                            int padLeft = Math.max(leftmostIndex-neededBases, 1);

                            int padRight = Math.min(leftmostIndex+reference.length+neededBases, referenceReader.getSequenceDictionary().getSequence(currentInterval.getContig()).getSequenceLength());

                            reference = referenceReader.getSubsequenceAt(currentInterval.getContig(), padLeft, padRight).getBases();

                            leftmostIndex = padLeft;

                        }

                        // now, fix the attribute tags

                        // TODO -- get rid of this try block when Picard does the right thing for reads aligned off the end of the reference

                        try {

                            if ( read.getAttribute(SAMTag.NM.name()) != null )

                                read.setAttribute(SAMTag.NM.name(), SequenceUtil.calculateSamNmTag(read, reference, leftmostIndex - 1));

                            if ( read.getAttribute(SAMTag.UQ.name()) != null )

                                read.setAttribute(SAMTag.UQ.name(), SequenceUtil.sumQualitiesOfMismatches(read, reference, leftmostIndex-1));

                        } catch (Exception e) {

                            // ignore it

                        }

                        // TODO -- this is only temporary until Tim adds code to recalculate this value

                        if ( read.getAttribute(SAMTag.MD.name()) != null )

                            read.setAttribute(SAMTag.MD.name(), null);

                        // mark that it was actually cleaned

                        readsActuallyCleaned.add(read);

                    }

                }

            }

            // END IF ( improvement >= LOD_THRESHOLD )

        } else if ( statsOutput != null ) {

            try {

                statsOutput.write(String.format("%s\tFAIL\t%.1f%n",

                        currentInterval.toString(), improvement));

                statsOutput.flush();

            } catch (Exception e) {

                throw new UserException.CouldNotCreateOutputFile("statsOutput", "Failed to write stats output file", e);

            }

        }

    }

    private void generateAlternateConsensesFromKnownIndels(final Set<Consensus> altConsensesToPopulate, final int leftmostIndex, final byte[] reference) {

        for ( VariantContext knownIndel : knownIndelsToTry ) {

            if ( knownIndel == null || !knownIndel.isIndel() || knownIndel.isComplexIndel() )

                continue;

            final byte[] indelStr;

            if ( knownIndel.isSimpleInsertion() ) {

                final byte[] fullAllele = knownIndel.getAlternateAllele(0).getBases();

                indelStr = Arrays.copyOfRange(fullAllele, 1, fullAllele.length); // remove ref padding

            } else {

                indelStr = Utils.dupBytes((byte)'-', knownIndel.getReference().length() - 1);

            }

            int start = knownIndel.getStart() - leftmostIndex + 1;

            Consensus c = createAlternateConsensus(start, reference, indelStr, knownIndel);

            if ( c != null )

                altConsensesToPopulate.add(c);

        }

    }

    private long determineReadsThatNeedCleaning(final List<GATKSAMRecord> reads,

                                                final ArrayList<GATKSAMRecord> refReadsToPopulate,

                                                final ArrayList<AlignedRead> altReadsToPopulate,

                                                final LinkedList<AlignedRead> altAlignmentsToTest,

                                                final Set<Consensus> altConsenses,

                                                final int leftmostIndex,

                                                final byte[] reference) {

        long totalRawMismatchSum = 0L;

        for ( final GATKSAMRecord read : reads ) {

            // we can not deal with screwy records

            if ( read.getCigar().numCigarElements() == 0 ) {

                refReadsToPopulate.add(read);

                continue;

            }

            final AlignedRead aRead = new AlignedRead(read);

            // first, move existing indels (for 1 indel reads only) to leftmost position within identical sequence

            int numBlocks = AlignmentUtils.getNumAlignmentBlocks(read);

            if ( numBlocks == 2 ) {

                Cigar newCigar = AlignmentUtils.leftAlignIndel(unclipCigar(read.getCigar()), reference, read.getReadBases(), read.getAlignmentStart()-leftmostIndex, 0, true);

                aRead.setCigar(newCigar, false);

            }

            final int startOnRef = read.getAlignmentStart()-leftmostIndex;

            final int rawMismatchScore = mismatchQualitySumIgnoreCigar(aRead, reference, startOnRef, Integer.MAX_VALUE);

            // if this doesn't match perfectly to the reference, let's try to clean it

            if ( rawMismatchScore > 0 ) {

                altReadsToPopulate.add(aRead);

                //logger.debug("Adding " + read.getReadName() + " with raw mismatch score " + rawMismatchScore + " to non-ref reads");

                if ( !read.getDuplicateReadFlag() )

                    totalRawMismatchSum += rawMismatchScore;

                aRead.setMismatchScoreToReference(rawMismatchScore);

                aRead.setAlignerMismatchScore(AlignmentUtils.mismatchingQualities(aRead.getRead(), reference, startOnRef));

                // if it has an indel, let's see if that's the best consensus

                if ( consensusModel != ConsensusDeterminationModel.KNOWNS_ONLY && numBlocks == 2 )  {

                    Consensus c = createAlternateConsensus(startOnRef, aRead.getCigar(), reference, aRead.getReadBases());

                    if ( c != null )

                        altConsenses.add(c);

                } else {

                    altAlignmentsToTest.add(aRead);

                }

            }

            // otherwise, we can emit it as is

            else {

                //logger.debug("Adding " + read.getReadName() + " with raw mismatch score " + rawMismatchScore + " to ref reads");

                refReadsToPopulate.add(read);

            }

        }

        return totalRawMismatchSum;

    }

    private void generateAlternateConsensesFromReads(final LinkedList<AlignedRead> altAlignmentsToTest,

                                                     final Set<Consensus> altConsensesToPopulate,

                                                     final byte[] reference,

                                                     final int leftmostIndex) {

        // if we are under the limit, use all reads to generate alternate consenses

        if ( altAlignmentsToTest.size() <= MAX_READS_FOR_CONSENSUSES ) {

            for ( AlignedRead aRead : altAlignmentsToTest ) {

                if ( CHECKEARLY ) createAndAddAlternateConsensus1(aRead, altConsensesToPopulate, reference,leftmostIndex);

                else createAndAddAlternateConsensus(aRead.getReadBases(), altConsensesToPopulate, reference);

            }

        }

        // otherwise, choose reads for alternate consenses randomly

        else {

            int readsSeen = 0;

            while ( readsSeen++ < MAX_READS_FOR_CONSENSUSES && altConsensesToPopulate.size() <= MAX_CONSENSUSES) {

                int index = GenomeAnalysisEngine.getRandomGenerator().nextInt(altAlignmentsToTest.size());

                AlignedRead aRead = altAlignmentsToTest.remove(index);

                if ( CHECKEARLY ) createAndAddAlternateConsensus1(aRead, altConsensesToPopulate, reference,leftmostIndex);

                else createAndAddAlternateConsensus(aRead.getReadBases(), altConsensesToPopulate, reference);

            }

        }

    }

    private void createAndAddAlternateConsensus(final byte[] read, final Set<Consensus> altConsensesToPopulate, final byte[] reference) {

        // do a pairwise alignment against the reference

         SWPairwiseAlignment swConsensus = new SWPairwiseAlignment(reference, read, swParameters);

         Consensus c = createAlternateConsensus(swConsensus.getAlignmentStart2wrt1(), swConsensus.getCigar(), reference, read);

         if ( c != null )

             altConsensesToPopulate.add(c);

    }

    private void createAndAddAlternateConsensus1(AlignedRead read, final Set<Consensus> altConsensesToPopulate,

                                                 final byte[] reference, final int leftmostIndex) {

         for ( Consensus known : altConsensesToPopulate ) {

              Pair<Integer, Integer> altAlignment = findBestOffset(known.str, read, leftmostIndex);

              // the mismatch score is the min of its alignment vs. the reference and vs. the alternate

              int myScore = altAlignment.second;

              if ( myScore == 0 ) {exactMatchesFound++; return; }// read matches perfectly to a known alt consensus - no need to run SW, we already know the answer

         }

         // do a pairwise alignment against the reference

         SWalignmentRuns++;

         SWPairwiseAlignment swConsensus = new SWPairwiseAlignment(reference, read.getReadBases(), swParameters);

         Consensus c = createAlternateConsensus(swConsensus.getAlignmentStart2wrt1(), swConsensus.getCigar(), reference, read.getReadBases());

         if ( c != null ) {

             altConsensesToPopulate.add(c);

             SWalignmentSuccess++;

         }

    }

    // create a Consensus from cigar/read strings which originate somewhere on the reference

    private Consensus createAlternateConsensus(final int indexOnRef, final Cigar c, final byte[] reference, final byte[] readStr) {

        if ( indexOnRef < 0 )

            return null;

        // if there are no indels, we do not need this consensus, can abort early:

        if ( c.numCigarElements() == 1 && c.getCigarElement(0).getOperator() == CigarOperator.M ) return null;

        // create the new consensus

        ArrayList<CigarElement> elements = new ArrayList<CigarElement>(c.numCigarElements()-1);

        StringBuilder sb = new StringBuilder();

        for (int i = 0; i < indexOnRef; i++)

            sb.append((char)reference[i]);

        int indelCount = 0;

        int altIdx = 0;

        int refIdx = indexOnRef;

        boolean ok_flag = true;

        for ( int i = 0 ; i < c.numCigarElements() ; i++ ) {

            CigarElement ce = c.getCigarElement(i);

            int elementLength = ce.getLength();

            switch( ce.getOperator() ) {

            case D:

                refIdx += elementLength;

                indelCount++;

                elements.add(ce);

                break;

            case M:

            case EQ:

            case X:

                    altIdx += elementLength;

            case N:

                if ( reference.length < refIdx + elementLength )

                    ok_flag = false;

                else  {

                    for (int j = 0; j < elementLength; j++)

                        sb.append((char)reference[refIdx+j]);

                }

                refIdx += elementLength;

                elements.add(new CigarElement(elementLength, CigarOperator.M));

                break;

            case I:

                for (int j = 0; j < elementLength; j++) {

                    if ( ! BaseUtils.isRegularBase(readStr[altIdx+j]) ) {

                        // Insertions with N's in them cause real problems sometimes; it's better to drop them altogether

                        ok_flag=false;

                        break;

                    }

                    sb.append((char)readStr[altIdx + j]);

                }

                altIdx += elementLength;

                indelCount++;

                elements.add(ce);

                break;

            case S:

            default:

                break;

            }

        }

        // make sure that there is at most only a single indel and it aligns appropriately!

        if ( !ok_flag || indelCount != 1 || reference.length < refIdx )

            return null;

        for (int i = refIdx; i < reference.length; i++)

            sb.append((char)reference[i]);

        byte[] altConsensus =  StringUtil.stringToBytes(sb.toString()); // alternative consensus sequence we just built from the current read

        return new Consensus(altConsensus, new Cigar(elements), indexOnRef);

    }

    // create a Consensus from just the indel string that falls on the reference

    private Consensus createAlternateConsensus(final int indexOnRef, final byte[] reference, final byte[] indelStr, final VariantContext indel) {

        if ( indexOnRef < 0 || indexOnRef >= reference.length )

            return null;

        // create the new consensus

        StringBuilder sb = new StringBuilder();

        Cigar cigar = new Cigar();

        int refIdx;

        for (refIdx = 0; refIdx < indexOnRef; refIdx++)

            sb.append((char)reference[refIdx]);

        if ( indexOnRef > 0 )

            cigar.add(new CigarElement(indexOnRef, CigarOperator.M));

        if ( indel.isSimpleDeletion() ) {

            refIdx += indelStr.length;

            cigar.add(new CigarElement(indelStr.length, CigarOperator.D));

        }

        else if ( indel.isSimpleInsertion() ) {

            for ( byte b : indelStr )

                sb.append((char)b);

            cigar.add(new CigarElement(indelStr.length, CigarOperator.I));

        } else {

            throw new IllegalStateException("Creating an alternate consensus from a complex indel is not allows");

        }

        if ( reference.length - refIdx > 0 )

            cigar.add(new CigarElement(reference.length - refIdx, CigarOperator.M));