1. A fix for the method to force the use of specific file formats (made by Patrick Lacasse)
2. Addition of a class to handle sequence annotation files in BIOML format (saxmodhandler).
3. Addition of a method to load the annotation file information into an STL map, in the class mprocess.
4. When compiling on Linux platforms, several possible makefiles are provided. The default makefile will work for GCC version 4, with the expat libraries dynamically linked. The other makefiles are all in the src directory, with names like "Makefile_XXX" where XXX is a descriptive name indicating in which situations this file is appropriate. To use these files, use a command line like this:
   

   	>make -f Makefile_GCCv3

1. An new override of the dtohl method in the saxhandler.cpp file was added to deal with 64 bit floats.

1. Including SAPS required modification of the classes mprocess and mscore as well as the addition of a new classs mscoresap, which is specified in the mscorepam.h file. The new class follows the same pattern as the other state machines for tracking sequence modifications. A class that reads the XML-formatted SAPS annotation information has also been added, saxsaphandler, which follows the same pattern as the other BIOML processing classes.
2. This version (and all subsequent ones) will use the preprocessor commands associated with the compiler make processor to specify the platform being compiled. Previous versions required the commenting out undesired options in the stdafx.h file. This change includes the PLUGGABLE_SCORING preprocessor definition in mscore.h that is necessary to alter the peptide scoring portion of the code.
3. The Mac OSX version of the executable binaries is statically linked to the most recent version of the XML parser expat. Unlike previous versions it will not be necessary to have expat installed on the computer used to run the search engine. The Linux/Unix builds are now the only platforms onwhich dynamic linking is necessary.

1. The output file has a new output value in the "performance parameters" group, for example:

   <group label="performance parameters" type="parameters">
       <note label="quality values">117 34 22 10 5 2 1 0 0 1 0 1 0 0 0 0 0 0 0 0</note>
   </group>

   
   This change was necessary to support the use of rho-diagrams in the GPM display software.
2. The proteins reported as possible correct identifications have been changed somewhat. In previous versions, it was possible for a protein to be reported as identified even if it did not have any qualifying peptides that were found to have the specified enzymatic cleavage: a protein could have peptides found only during the cleavage-at-every site round or the point-mutation round. This behavior has been changed so that a protein must have at least one significant peptide found to have the specified enzymatic cleavage.
3. The implementation of point mutation detection has been altered so that if a particular possible point mutation has been explained by any set of potential modifications, it will not be included as a possible solution in the output.
4. The stdafx.h file has been altered to add in new preprocessor statements that deal with the different versions of Mac OS X. These new statements are #OSX_TIGER and #OSX_INTEL. To compile with OS X 10.4 on a PPC computer, uncomment both #OSX and #OSX_TIGER. To compile with OS X 10.4 on an Intel computer, uncomment only #OSX_INTEL.

1. Code that uses iterator math to determine the limits of a calculation have been altered so that the iterators are not incremented passed the end of an STL container. Incrementing iterators passed the end of a container generates a run-time error when compiled with Microsoft Visual C++ 2005. The precompiler variable _CRT_SECURE_NO_DEPRECATE has been defined for the Visual C++ compiler, to prevent the generation of unnecessary compiler warnings for the use of C string functions, such as "strcpy".
2. The SAXTandemInputHandler::characters method has been updated to improve its performance and to handle escape characters correctly (suggested by Brendan Maclean).
3. The documentation and precompiler defines in stdafx.h that provide cross-platform compatibility for 64-bit integer types have been updated.

1. On at least some Linux platforms, astersiks (*) in FASTA file were not being processed properly. This has been corrected by Brendan Maclean.
2. Some mzXML files could produce memory problems, when combined with some spectrum processing parameters because a vector was not being cleared between processing individual spectra. This problem did not affect the results of the search, but it could cause memory paging when using a large file.
3. The taxonomy xml file has always contained type specification, which was not used by X! Tandem. Now X! Tandem enforces that FASTA or FASTA.PRO files must be specificed with format="peptide".

This release of X! Tandem includes a number of additions to the system API. These changes are mainly for programmers, allowing for greater customization of searches. Some of these features have been present in previous versions of X! Tandem, but have been either undocumented or unsufficiently well tested.

This version of the X! Tandem code also merges the code for X! P3. The P3 executable can be compiled by uncommenting the preprocessor variable X_P3 in "stdafx.h". Adding in the P3 code was done by creating several classes that are extensions of the normal Tandem classes and using a small number of preprocessor directives. These new classes have the prefix p3.

1. output, log path
2. output, message
3. output, one sequence copy
4. output, sequence path
5. refine, modification mass
6. refine, sequence path
7. refine, tic percent
8. scoring, cyclic permuation
9. scoring, include reverse
10. spectrum, sequence batch size

This release of X! Tandem includes a number of additions to the system API. These changes are mainly for programmers, allowing for greater customization of searches. Some of these features have been present in previous versions of X! Tandem, but have been either undocumented or unsufficiently well tested.

This version of the X! Tandem code also merges the code for X! P3. The P3 executable can be compiled by uncommenting the preprocessor variable X_P3 in "stdafx.h". Adding in the P3 code was done by creating several classes that are extensions of the normal Tandem classes and using a small number of preprocessor directives. These new classes have the prefix p3.

1. output, log path
2. output, message
3. output, one sequence copy
4. output, sequence path
5. refine, modification mass
6. refine, sequence path
7. refine, tic percent
8. scoring, cyclic permutation
9. scoring, include reverse
10. spectrum, sequence batch size

1. The C-ion mass calculation has been improved for electron-capture ion source identifications, suggested by David Fenyo.
2. A problem relating to an include file that caused compilation difficulties for some versions of GCC on some version of Linux has been fixed.
3. The calculation of parent ion mass difference has been improved, to provide better consistency for very accurate (< 1 ppm) parent ion mass determinations.
4. The "semi" cleavage state machine has been adjusted for better performance.

1. An improved handling of hex encoded binary information in mzXML and mzData files, for 64-bit processors, added by Steven Wiley.
2. Addition of testing for N-terminal glutamic acid cyclization, suggested by Oleg Krohkin.
3. Addition of "semi" enzymatic cleavage (specific enzyme cleavage at one end of a peptide and non-specific cleavage at the other), suggested by Matt Monroe.
4. An improved system for detecting XML file types, suggested by Steven Wiley.
5. Support for variant methods of expressing parent ion charge in mzData v. 1.05, added by Fredrik Levander.

1. An improved version of the state machine that lists all of the possible potential modification states of a peptide sequence was written by Brendan Maclean. This version is both more thorough and faster than the previous code.
2. The capability of using chemical average masses for fragment ion mass calculations was added by Brendan Maclean.
3. A simplification of the mprocess class that allows for a "pluggin" approach to adding new refinement modules was designed and implemented by Rob Craig.
4. An improved routine for correcting for isotope peaks and multiple observations of similar masses was made by Patrick Lacasse.
5. An additional state machine using cyclic peptide sequence permutation to compensate for small sequence collections and for large mass peptides was added. This feature is based on a suggestions made separately by Tom Blackwell & David States and Patrick Lacasse.
6. An improved sorting method to improve the consistency of homologous sequence assignments was added by Rob Craig.

1. The handlers for GAML spectra, taxonomy files and input parameter files have to changed to using expat, rather than custom routines.
2. A more flexible mass calculation class has been added to improve molecular mass consistency for high accuracy calculations.
3. The input spectrum file type detection method has been improved by adding the possibility of forcing it to select one file type. This forcing is done using the input parameter "spectrum, path type" parameters, which can have the values: dta, pkl, mgf, gaml, mzxml or mzdata. If this parameter is missing or of zero length, the normal file type detection scheme is used.

1. New classes have been added to allow the processing of mzxml and mzdata file formats.
   Two of the new classes are publicly derived from loadspectrum, a custom class specific to Tandem. Two others are publicly derived from the xml parser class SAXSpectraHandler which is imported from the expat library of functions. The xml parser classes use the expat functions exclusively to parse the input in order to load it into the traditional Tandem spectra data members.
2. base64.cpp and base64.h have been added to allow b64_decode_mio() function calls, which are needed to decode the spectra in mzxml and mzdata spectra files.
3. Included in the src folder is the libexpat.lib which is required to compile new versions of the executable on Windows. Linux and OSX machines should have the required libraries as part of the core operating system.
4. A new function has been added to msequtilites that allows amino acid residue masses to be defined by an xml input file. If the parameter 'protein, modified residue mass file' is defined in the input.xml, the masses are taken from the file defined by that parameter. An example of the format can be viewed here.

1. Several new classes have been added, to make the scoring system "pluggable", i.e., it is now much easier to alter the scoring system used, for the purposes of bioinformatics investigations. These changes are mainly of interest to informatics professionals and they should not affect the normal operation of the software for users.
2. The calculation of parent ion mass has been changed, taking more care as to the mass of added groups and correctly accounting for electron masses.
3. Better statistical methods have been added to deal with the small number of possible peptides generated from a list of protein sequences that have a very high accurately determined parent ion mass.

1. Spectra that are interpreted as being caused by a prompt neutral loss now have the prompt loss specified in the appropriate <aa> node in the output.
2. Correction of an issue with the OS X version that resulted in improper reading of ".pro" sequence files. Initially, the ".pro" format was to have both little endian and big endian versions, however this became too confusing to maintain. The current plan is to only use the little endian format and to compensate for this on-the-fly in the OS X version.
3. The maximum parent ion charge to be used can now be specified using the "spectrum, maximum parent charge" parameter. This parameter has a default value of 4. This change was made necessary because of high charge states being called by some MS peak assignment software, which caused spurious assignments.
4. The first round of refinement (finding partially cleaved peptides) has been extended, so that it possible to repeat it with different sets of modifications and motifs. These additional refinement rounds are specified by adding parameters using the following format:
   • Round 1: "refine, potential modification mass"
     "refine, potential modification motif"
   • Round 2: "refine, potential modification mass 1"
     "refine, potential modification motif 1"
   • Round 3: "refine, potential modification mass 2"
     "refine, potential modification motif 2"
   This will continue until both of the next pair of parameters are either missing or neither contain an ampersand (@).

1. An error that resulted in the incorrect interpretation of some motifs was corrected.
2. Addition of spectrum prefiltering to remove repeated spectra from the initial set of mass spectra. This feature compares spectra using a dot product calculation and removes spectra that have vector representations that point in the same direction. The most intense spectrum out of a set of repeated spectra is kept and used for analysis. This type of filtering can remove up to 90% of spectra from a MudPit-style run, making data analysis and interpretation easier.

1. A possible floating-point overrun error that could lead to 0.0 expectation values for high scoring peptides was detected from GPMDB submissions has been corrected.

1. An error that reduced the score for triply-charged ions was corrected.
2. Quantitation information was added to the output XML file.

1. The memory management throughout the program has been analyzed and altered to minimize the amount of memory used per spectrum. This effort has reduced the amount of memory used in single threaded operation by as much as 60%: the improvement for double threaded operation may be as much as 80%.
2. The threading model has been changed to allow for the use of multiple processeors in the refinement process. TANDEM 1 separated work between the threads by dividing up the sequences to search, so that each thread would only search a subset of the sequences in a FASTA file. TANDEM 2 divides up the mass spectra between threads, so that each thread searches a subset of the mass spectra. This change makes it easier to divide up the refinement job, but means that running more than one thread on a single processor will degrade the performance of the software. For best performance, it is now important to keep the number of threads and the number of processors the same.
3. The refinement process has been improved in accuracy by applying a logical filter after each step of refinement. This means that once a refinement step is completed, the new results obtained from the refinement are examined and it the new results are not significantly better than those obtained from a simpler search, they are discarded and the simpler results retained. This filtering significantly reduces the complexity of analyzing results when there may be a variety of similar modification patterns or point mutations that explain a particular spectrum.
4. Validation of results using reversed sequence databases has been built-in to the search process. This validation may be turned on or off, using the new input parameter "scoring, include reverse" (values = yes|no). This validation process tabulates the number of unique high probability hits from the reversed sequence search and places them in the output file, along with estimates of the false positive rate based on TANDEM's stochastic histogramming technique and the estimate derived from the reversed sequence process. NOTE: When this validation method is used, twice as many sequences must be processed (both forward and reversed), which may require significantly more processing time.
5. Numerous small optimizations have been made, particularly for loading and reporting the results for very large collections of mass spectra.

1. The ability to specify modifications based on sequence motifs was added to both the normal search and refinement steps. A comma separated list of motifs in slightly modified PROSITE format can be used to only modify specific residues. An example of this format is:
   204@[N!]{P}[ST]{P} - which says a motif that has an N, followed by any residue except P, followed by an S or a T, followed by any residue except P is specified. Modify the residue in the group containing the exclamation point (in this case the N) by adding 204 Da.
   The peptides containing this motif are checked both with and without this modification, so it is interpreted the same way that a "potential" modification is interpretted. The rules for creating these motifs are:
   • Square brackets "[]" indicate any of the residues contained is possible;
   • French brackets "{}" indicate that any of the residues contained is forbidden;
   • A bare letter is interpreted as if it was in square brackets and can be modified with an exclamation point, e.g. 16@[M!] is the same as 16M!;
   • An exclamation point indicates the position of modification;
   • The letter "X" indicates any residue;
   and
   • Round brackets "()" indicate a count, e.g. "X(10)" means ten X's in a row;
   • All other characters are ignored, e.g. 80@[ST!]PX[KR] is the same as 80@[ST!]-P-X-[KR].
2. The ability to specify prompt neutral losses for potential modifications (including motifs) has been added. This neutral loss is specified by adding a colon followed by the mass corresponding to the loss. For example:
   80@S specified phosporylation without loss, while 80:-98@S specifies the neutral loss of the phosphate group.
3. The ability to specify on a sequence by sequence level specific fixed modifications of residues by the residue number has been added. This capability cannot be exploited currently because of a lack of sequence lists that contain this type of information. However, an appropriately translated version of a database such as SWISSPROT could be used to provide this information.

1. The mechanism for storing sequences in the mprocess class has been changed. Previously, a copy of each protein sequence was stored with each peptide model associated with a spectrum. Now, a master list of protein sequences is kept and only a lookup number is stored with each peptide model. This change improves memory management for very large pools of redundant proteins or very long lists of spectra.

1. The refinement processing was returned to its previous state, so that only the best scoring peptides from the refinement process are reported.

1. GAML spectra, such as those in output xml files, can now be used as input data.
2. The length of scoring histogram arrays have been altered to improve memory usage.
3. Several instances of temporary copies of data have been removed and other data structures cleared as soon as possible after use.

1. A behavior that resulted in the lost of the last character in sequences in some FASTA files has been corrected.
2. A compatibility issue resulting from various choices for the size of the size_t STL variable on unix platforms has been corrected, so that most unix platforms should compile without modifying the linux version of the code.

1. No problems known at time of release

1. A new state machine was added to perform N- and C- terminal partial modifications. Previous versions used these modifications as complete modifications only.
2. The optimization for the minimum number of residues considered was removed and replaced with the constant value of 4. The prior optimization did not produce a significant improvement in speed, but it did cause occasional problems with large neutral losses.
3. The xslt and css files have been updated to conform more closely to specification, making them compatible with the FireFox browser.

1. A behavior that allowed the occasional consideration of peptides with too many missed cleavage sites was fixed.
2. A compatibility issue for starting threads on some unix platforms has been corrected, so that most unix platforms should compile without modifying the linux version of the code.

1. No problems known at time of release

1. A new state machine was added to the mscore object to track point mutations.
2. A new report value was added so that a protein sequence would only recorded once in the XML output, if desired.

1. Several problems associated with detected unsupported spectrum files types were corrected. Previous version could hang indefinately if binary files were used.
2. A LINUX compilation problem with some flavours of LINUX that caused a failure to create new threads was corrected.

1. No problems known at time of release

1. A new statistical interpretation was added, to combine expectation values when multiple models from the same sequence are found to be the best model in different spectra. Using this model, expectation values for the collections of models are now listed as the base-10 log of the expectation value, beside the FASTA description line.
2. The way FASTA description lines are listed has been changed. Rather than listing the descriptions in the same order they were encountered in the search, they are now listed by length: the longest entry first. The logic to this choice is that for the NCBI database nr, the oldest entry for a similar sequence tends to be the longest and the first line of that entry tends to have the best description of the protein's common name. Unfortunately, this is not always true.
3. A new way of organizing the output was added. It can be accessed by setting the output, sort results by parameter value to protein. Models corresponding to a given sequence are grouped together, with the best set of models at the top of the page.

1. FASTA file name problem fixed.
2. Multiple modification reporting problem fixed.

1. No problems known at time of release

1. A threading model was introduced that allows up to 16 threads. Each thread is given a unique set of sequences to model and the results of all of the models are summed at the end. The threads are started in a simple manner and then the program waits for all threads to return before summing and correlating the data.
2. A new class called "mspectrumcondition" was added to perform any spectrum filtering necessary. The initial release had this functionality in the "mscore" class.
3. An example XSLT and CSS stylesheet pair were added, that allow viewing the output XML with a browser somewhat easier. They also are an example of how such a pair of files can be constructed to create a GUI from the output XML.

1. The memory leak found in the 2003.05.01 release was found and fixed.

1. Some FASTA sequence list file names are not being recorded in the output XML.
2. Occasionally, multiple potential modifications are noted for the same residue.
3. The XSLT is not compatible with Internet Explorer 5.5. This is by design and it will not change in later releases.

1. Soon after the release, a serious memory leak was reported, which became evident when searching large data files. This leak reduced system performance dramatically.