Difference between revisions of "PhyloSoC:BioPerl integration of the NeXML exchange standard and Bio::Phylo toolkit/HOWTO"

From Phyloinformatics
Jump to: navigation, search
Line 4: Line 4:
[http://bio.perl.org/wiki/User:Chmille4 Chase Miller]
[http://bio.perl.org/wiki/User:Chmille4 Chase Miller]

Revision as of 15:26, 26 July 2009


This is a HOWTO about the Template:PM module, and how to use it to read and write complete Nexml documents. We will also describe how the Template:PM, Template:PM, and Template:PM modules work for outputting individual objects to nexml.


Chase Miller


These nexml modules integrate the NeXML exchange standard into BioPerl, facilitating the adoption of this standard and easing the transition from the overworked NEXUS standard. A wrapper was used to allow BioPerl native access to the preferred NeXML parser (Bio::Phylo), allowing Bio::Phylo and NeXML to co-evolve without being encumbered by BioPerl


The Template:PM system was designed with the following assumptions: That all reports parsed with it could be separated into a hierarchy of components. The Result is the entire analysis for a single query sequence, and multiple Results can be concatenated together into a single file (i.e. running blastall with a fasta database as the input file rather than a single sequence). Each Result is a set of Hits for the query sequence. Hits are sequences in the searched database which could be aligned to the query sequence and met the minimal search parameters, such as e-value threshold. Each Hit has one or more High-scoring segment Pairs (HSPs) which are the alignments of the query and hit sequence. Each Result has a set of one or more Hits and each Hit has a set of one or more HSPs, and this relationship can be used to describe results from all pairwise alignment programs including BLAST, FastA, and implementations of the Smith-Waterman and Needleman-Wunsch algorithms.

A design pattern, called Factory, is utilized in object oriented programming to separate the entity which processes data from objects which will hold the information produced. In the same manner that the Template:PM module is used to parse different file formats and produces objects which are Template:PM compliant, we have written Template:PM to produce the Bio::Search objects. Sequences are a little less complicated so there is only one primary object (Template:PM) while Search results need three main components to represent the data processed in a file:

The Template:PM object is then a factory which produces Template:PM objects that contain information about the query, the database searched, and the full collection of Hits found for the query.

The generality of the Template:PM approach is demonstrated by the large number of report formats that have appeared since its introduction. These formats are listed below.

Table 1: SearchIO modules and formats supported
Name Description
fasta FASTA -m9 and -m0
blasttable BLAST tabular -m9 or -m8 (NCBI) and -mformat 2 or -mformat 3 (WU-BLAST)
erpin ERPIN versions 4.2.5 and above
infernal Infernal versions 0.7 and above
megablast MEGABLAST
psl UCSC formats PSL
waba WABA
axt AXT
sim4 Sim4
hmmer HMMER hmmpfam and hmmsearch
exonerate Exonerate CIGAR
wise Genewise -genesf
rnamotif raw rnamotif output for RNAMotif versions 3.0 and above

Parsing with Bio::SearchIO

This section is going to describe how to use the Template:PM system to process reports. We'll describe BLAST reports but the idea is that once you understand the methods associated with the objects you won't need to know anything special about other Template:PM parsers.

Avoiding possible confusion

Before we get into the details we should admit that there is some confusion about the names and functions of the objects for historical reasons.

Both Steve Chervitz and Jason Stajich have implemented parsers in this system. The basic objects Jason has created are called Bio::Search::XXX::GenericXXX where, again, XXX is HSP, Hit, and Result. Most of the implementations use these simple objects for sorting the data. Steve created the psiblast parser (which was later merged into the Template:PM module) and a host of objects named Bio::Search::XXX::BlastXXX where XXX is HSP, Hit, and Result. These objects have additional functions related to output from BLAST.

The important take home message is that you cannot assume that methods in the BlastXXX objects are in fact implemented by the GenericHSP objects. More likely than not the BlastXXX objects will be deprecated and dismantled as their functionality is ported to the GenericHSP objects. For this reason we'll only be discussing the Generic* objects, though we'll use the terms 'hit', 'HSP', and 'result'.

Using SearchIO

HOWTO:SearchIO/Using In order to display all these methods and what they return let's use a report as input, a simple BLASTX result:

BLASTX 2.2.4 [Aug-26-2002]
Reference: Altschul, Stephen F., Thomas L. Madden, Alejandro A. Schaffer,
Jinghui Zhang, Zheng Zhang, Webb Miller, and David J. Lipman (1997),
"Gapped BLAST and PSI-BLAST: a new generation of protein database search
programs",  Nucleic Acids Res. 25:3389-3402.
Query= gi|20521485|dbj|AP004641.2 Oryza sativa (japonica
cultivar-group) genomic DNA, chromosome 1, BAC clone:B1147B04, 3785
bases, 977CE9AF checksum.
         (3059 letters)
Database: test.fa
           5 sequences; 1291 total letters

                                                                Score    E
Sequences producing significant alignments:                     (bits) Value
gb|443893|124775 LaForas sequence                                 92   2e-022
>gb|443893|124775 LaForas sequence
          Length = 331
 Score = 92.0 bits (227), Expect = 2e-022
 Identities = 46/52 (88%), Positives = 48/52 (91%)
 Frame = +1

  Database: test.fa
    Posted date:  Feb 12, 2003  9:51 AM
  Number of letters in database: 1291
  Number of sequences in database:  5
Lambda     K      H
   0.318    0.135    0.401
Lambda     K      H
   0.267   0.0410    0.140

Matrix: BLOSUM62
Gap Penalties: Existence: 11, Extension: 1
Number of Hits to DB: 7140
Number of Sequences: 5
Number of extensions: 180
Number of successful extensions: 2
Number of sequences better than 10.0: 2
Number of HSP's better than 10.0 without gapping: 1
Number of HSP's successfully gapped in prelim test: 0
Number of HSP's that attempted gapping in prelim test: 0
Number of HSP's gapped (non-prelim): 1
length of database: 1291
effective HSP length: 46
effective length of database: 1061
effective search space used:  1032353
frameshift window, decay const: 50,  0.1
T: 12
A: 40
X1: 16 ( 7.3 bits)
X2: 38 (14.6 bits)
X3: 64 (24.7 bits)
S1: 32 (17.6 bits)

NCBI-BLAST parsing problems

A plaintext NCBI-BLAST report like that used in the example above probably remains the most common BLAST output format in use. However, since the NCBI has stated that this format can change without warning, the SearchIO BLAST parser will break from time to time. To avoid this instability, you can have NCBI-BLAST produce its report in XML format with the -m7 command-line option. To parse blastxml reports with SearchIO, use -format=>'blastxml' instead of -format=>'blast'. Parsing BLAST XML output may also be faster. WU-BLAST offers XML output as well through the command-line option mformat=7, although its standard reports may work when NCBI-BLAST's do not.

Table of Methods

Object Method Example Description
Result algorithm BLASTX algorithm string
Result algorithm_version 2.2.4 [Aug-26-2002] algorithm version
Result query_name 20521485|dbj|AP004641.2 query name
Result query_accession AP004641.2 query accession
Result query_length 3059 query length
Result query_description Oryza sativa ... 977CE9AF checksum. query description
Result database_name test.fa database name
Result database_letters 1291 number of residues in database
Result database_entries 5 number of database entries
Result available_statistics effectivespaceused ... dbletters statistics used
Result available_parameters gapext matrix allowgaps gapopen parameters used
Result num_hits 1 number of hits
Result hits List of all Bio::Search::Hit::GenericHit object(s) for this Result
Result rewind Reset the internal iterator that dictates where next_hit() is pointing, useful for re-iterating through the list of hits.
Hit name 443893|124775 hit name
Hit length 331 Length of the Hit sequence
Hit accession 443893 accession (usually when this is a genbank formatted id this will be an accession number- the part after the gb or emb )
Hit description LaForas sequence hit description
Hit algorithm BLASTX algorithm
Hit raw_score 92 hit raw score
Hit significance 2e-022 hit significance
Hit bits 92.0 hit bits
Hit hsps List of all Bio::Search::HSP::GenericHSP object(s) for this Hit
Hit num_hsps 1 number of HSPs in hit
Hit locus 124775 locus name
Hit accession_number 443893 accession number
Hit rewind Resets the internal counter for next_hsp() so that the iterator will begin at the beginning of the list
HSP algorithm BLASTX algorithm
HSP evalue 2e-022 e-value
HSP expect 2e-022 alias for evalue()
HSP frac_identical 0.884615384615385 Fraction identical
HSP frac_conserved 0.923076923076923 fraction conserved (conservative and identical replacements aka "fraction similar")
(only valid for Protein alignments will be same as frac_identical)
HSP gaps 2 number of gaps
HSP query_string DMGRCSSG .. query string from alignment
HSP hit_string DIVQNSS ... hit string from alignment
HSP homology_string D+ + SSGCN ... string from alignment
HSP length('total') 52 length of HSP (including gaps)
HSP length('hit') 50 length of hit participating in alignment minus gaps
HSP length('query') 156 length of query participating in alignment minus gaps
HSP hsp_length 52 Length of the HSP (including gaps) alias for length('total')
HSP frame 0 $hsp->query->frame,$hsp->hit->frame
HSP num_conserved 48 number of conserved (conservative replacements, aka "similar") residues
HSP num_identical 46 number of identical residues
HSP rank 1 rank of HSP
HSP seq_inds('query','identical') (966,971,972,973,974,975 ...) identical positions as array
HSP seq_inds('query','conserved-not-identical') (967,969) conserved, but not identical positions as array
HSP seq_inds('query','conserved') (966,967,969,971,973,974,975, ...) conserved or identical positions as array
HSP seq_inds('hit','identical') (197,202,203,204,205, ...) identical positions as array
HSP seq_inds('hit','conserved-not-identical') (198,200) conserved not identical positions as array
HSP seq_inds('hit','conserved',1) (197,202-246) conserved or identical positions as array, with runs of consecutive numbers compressed
HSP score 227 score
HSP bits 92.0 score in bits
HSP range('query') (2896,3051) start and end as array
HSP range('hit') (197,246) start and end as array
HSP percent_identity 88.4615384615385  % identical
HSP strand('hit') 1 strand of the hit
HSP strand('query') 1 strand of the query
HSP start('query') 2896 start position from alignment
HSP end('query') 3051 end position from alignment
HSP start('hit') 197 start position from alignment
HSP end('hit') 246 end position from alignment
HSP matches('hit') (46,48) number of identical and conserved as array
HSP matches('query') (46,48) number of identical and conserved as array
HSP get_aln sequence alignment Template:PM object
HSP hsp_group Not available in this report Group field from WU-BLAST reports run with -topcomboN or -topcomboE specified
HSP links Not available in this report Links field from WU-BLAST reports run with -links showing consistent HSP linking
Table 2: All the data returned by methods used by the Result, Hit, and HSP objects when the report shown above is used as input. Note that many of the methods shown can be used to either get or set values, but we're just showing what they get.

Using the methods

Table 2 shows that a method can return a string, an array, or an object. When an object is returned some additional code will probably be needed to get the data of interest. For example, if you wanted a printable alignment after you'd parsed BLAST output you could use the get_aln() method, retrieve a Template:PM object and use it like this: <perl> use Bio::AlignIO;

  1. $aln will be a Bio::SimpleAlign object

my $aln = $hsp->get_aln; my $alnIO = Bio::AlignIO->new(-format =>"msf", -file => ">hsp.msf"); $alnIO->write_aln($aln); </perl> On one hand it appears to be a complication, but by entering the worlds of the Template:PM and Template:PM objects you now have access to their functionality and flexibility. This is the beauty of BioPerl!

Some of these methods deserve a bit more explanation since they do more than simply extract data directly from the output. For example, the ambiguous_aln() method is designed to tell us whether two or more HSPs from a given hit overlap, and whether the overlap refers to the queries or the hits, or both. One situation is where overlaps would be found in one but not the other arises where there are repeats in the query or hit. The ambiguous_aln() method will return one of these 4 values:

q query sequence contains overlapping sub-sequences while hit sequence does not
s hit sequence contains overlapping sub-sequences while query does not
qw query and hit sequences contain overlapping sub-sequences relative to each other
- query and hit sequence do not contain multiple domains relative to each other OR both contain the same distribution of similar domains

Another method that's useful in dissecting an HSP is the seq_inds() method of the HSP object. What this method does is tell us what the positions are of all the identical, conserved, mismatched, or gap ("identical", "conserved", "nomatch", "gap") residues in the query or hit sequence as deduced from the HSP alignment. The returned positions refer to the query or hit sequence ("sbjct" is synonymous with "hit"). It could be used like this: <perl>

  1. put all the conserved matches in query strand into an array

my @str_array = split "",$hsp->query_string; foreach ( $hsp->seq_inds('query','conserved') ){

 push @conserved,$str_array[$_ - 1];

} </perl>

For 'gaps', the returned positions are the sequence indices prior to a gap insertion; if the gap insert is greater than 1 then the position is repeated based on the number of insertions. You can regain the gap length using a simple temporary hash:


  1. grab hsp

my %gap_pos; for my $pos ($hsp->seq_inds('query'=>'gaps')) {


} print "Ind: $_ Gaps: $gap_pos{$_}\n" for sort {$a <=> $b} keys %gap_pos; </perl>

seq_inds() can be very useful for extracting the mismatch bases in an alignment. If you wanted to figure out which bases are not matching in an alignment you could use seq_inds to get these positions and then extract out these specific bases from the alignment. Note: translated sequences in a HSP (such as those used for TBLASTN, BLASTX, etc.) are by nature ambiguous as the sequence coordinates reported in the HSP map back to the original (nucleotide) sequence; under these circumstances all redundant nucleotide positions are returned for that match.

One final note when using seq_inds(): if you want a list of ranges or only care about the positions of the gaps (not the gap length) and don't want repeated positions, you can use my @inds = $hsp->seq_inds('query'=>'gaps',1); the last argument is a boolean flag with collapses consecutive positions into ranges and single sequence positions.

In most cases the SearchIO methods extract data directly from output but there's one important exception, the frame() method of the HSP object. Instead of using the values in the BLAST report it converts them to values according to the GFF specification, which is a format used by many BioPerl modules involved in gene annotation.

Specifically, the frame() method returns 0, 1, or 2 instead of the expected -3, -2, -1, +1, +2, or +3 in BLAST. GFF frame values are meaningful relative to the strand of the hit or query sequence so in order to reconstruct the BLAST frame you need both the strand, 1 or -1, and the GFF frame value: <perl>

my $blast_frame = ($hsp->query->frame + 1) * $hsp->query->strand;

</perl> Another common example is analyzing data from a single BLAST report several different ways, such as sorting hits based on particular criteria then printing all hits and HSPs. One could iterate through the list twice, once for sorting the objects and then a second time for printing. Note that in this script:

  • A Template:PM object is passed to the subroutines, not the SearchIO object in $blast_report
  • You must use rewind to reset the iterator for the Template:PM object since there are two iterations through this object (one in each subroutine)

<perl> ... my $result = $blast_report->next_result; &sort_results($result); $result->rewind; &print_blast_results($result);

sub sort_results{

  my $result = shift;
  my @hits;
  while(my $hit = $result->next_hit()){
     push @hits, $hit;
  # sort by accessions
  my @acc = sort {$a->accession cmp $b->accession} @hits;
  print join("\t", $_->accession,$_->description),"\n" for @acc;


sub print_blast_results{

  my $result = shift;
  while(my $hit = $result->next_hit()){
     while(my $hsp = $hit->next_hsp()){
        print join(", ",$hit->name,$hsp->bits)."\n";

} </perl> If you parsed a report already and want to reset the parser (i.e. if you sent the SearchIO object to the subs above instead of the Template:PM object, then iterated through everything twice), you would need to reset the SearchIO object itself by using seek($blast_report->_fh, 0);. We don't recommend doing this for two reasons. First, each round of parsing takes the same length of time since you start parsing the report from scratch, so you'll take a time hit. Second, if you saved objects from the first round of parsing you will take a memory hit, because a new set of objects is generated for each subsequent round of parsing the report. Hence, we use the already-generated Template:PM object in the subroutines instead.

To simplify things here a bit more, you could simply grab all the hits at once using $result->hits before you sort them, as demonstrated in a previous example. The hits method does not use the iterator; it simply returns a list of BlastHit objects. Hence, there is no need to rewind the Template:PM object:

<perl> ... &sort_results($result); &print_blast_results($result);

sub sort_results{

  my $result = shift;
  my @acc = sort {$a->accession cmp $b->accession} $result->hits;
  print join("\t", $_->accession,$_->description),"\n" for @acc;

} </perl>

Similar methods exist for Template:PM objects to rewind the iterator for HSP's ($hit->rewind) and to grab all HSP's($hit->hsps).

Our simple table of methods does not show all available arguments or returned values for all the SearchIO methods. The best place to explore any method in detail is http://doc.bioperl.org which provides the HTML versions of the Perl POD (Plain Old Documentation) that is embedded in every well-written Perl module - there's also a list of modules at the bottom of this HOWTO. Another source of examples is the Bioperl scripts which come with the BioPerl packages.


One frequently-asked question has to do with getting sorted output from a report, or sorting hits or HSPs just as they're sorted in the input file. There's little in the way of sorting in Template:PM's methods, generally speaking you'll just use a standard Perl approach. Here is an example that sorts hits according to their bit scores: <perl>

my @hits = $result->hits;
for my $hit ( sort { $a-> bits <=> $b->bits } @hits ) {
  # Do something...


Creating Reports for SearchIO

One note on creating reports that can be parsed by Template:PM: the developers haven't attempted to parse all the possible reports that could be created by programs with many command-line options, like blastall. Certainly you should be able to parse reports created using the default settings, but if you're running blastall, say, using some special set of options and you've encountered a parsing problem this may be the explanation.

For example, one can currently parse BLAST output created with the default settings as well as the reports created when using the "-m 8" or "-m 9" options (use -format=>'blasttable') or the "-m 7" XML-formatted reports (use -format=>'blastxml') but it's still possible to find sets of options that Template:PM can't parse.

You might also find it useful not to have to create reports as files. Template:PM, like Template:PM, is aware of STDIN so you can pipe output from the search application directly to it (on operating systems that allow such things). It could look something like this: <perl> use strict; use Bio::SearchIO;

my $fh; my $fasta = "/usr/local/bin/fasta34"; my $library = "hs.seq"; my $query = "deserts.seq"; my $options = "-E 0.01 -m 0 -d 10 -Q"; my $command = "$fasta $options $query $library";

open $fh,"$command |" || die("cannot run fasta cmd of $command: $!\n");

my $searchio = Bio::SearchIO->new(-format => 'fasta', -fh => $fh); </perl>


This section is going to describe how the SearchIO system was implemented, it is probably not necessary to understand all of this unless you are curious or want to implement your own Template:PM parser. We have utilized an event-based system to process these reports. This is analogous to the SAX (Simple API for XML) system used to process XML documents. Event based parsing can be simply thought of as simple start and end events. When you hit the beginning of a report a start event is thrown, when you hit the end of the report an end event is thrown. So the report events are paired, and everything else that is thrown in between the paired start and end events is related to that report.

Another way to think of it is as if you pick a number and color for a card in a standard deck. Let's say you pick red and 2. Then you start dealing cards from our deck and pile them one on top of each other. When you see your first red 2 you start a new pile, and start dealing cards onto that pile until you see the next red 2. Everything in your pile that happened between when you saw the beginning red 2 and ending red 2 is data you'll want to keep and process. In the same way all the events you see between a pair of start and end events (like 'report' or 'hsp') are data associated with object or child object in its hierarchy. A listener object processes all of these events, in our example the listener is the table where the stack of cards is sitting, and later it is the hand which moves the pile of cards when a new stack is started. The listener will take the events and process them. We've neglected to tell you of a third event that is thrown and caught. This is the characters event in SAX terminology, which is simply data. So one sends a start event, then some data, then an end event. This process is analogous to a finite state machine in computer science where what we do with data received is dependent on the state we're in. The state that the listener is in is affected by the events that are processed.

A small caveat: in an ideal situation a processor would throw events and not need to maintain any state information, it would just be processing data and the listener would manage the information and state. However, a lot of the parsing of these human-readable reports requires contextual information to apply the correct regular expressions. So in fact the event thrower has to know what state it is in and apply different methods based on this. In contrast the XML parsers simply keep track of what state they are in, but can process all the data with the same system of reading the tag and sending the data that is in between the XML start and end tags.

All of this framework has been built up, so to implement a new parser one only needs to write a module that produces the appropriate start and end events and the existing framework will do the work of creating the objects for you. Here's how we've implemented event-based parsing for Template:PM. The Template:PM is just the front-end to this process, in fact the processing of these reports is done by different modules in the Bio/SearchIO/ directory. So if you look at your BioPerl distribution and the modules in Bio/SearchIO you'll see modules in there like blast.pm, fasta.pm, blastxml.pm, SearchResultEventBuilder.pm, EventHandlerI.pm (depending on what version of the toolkit there may be more modules in there). There is also a SearchWriterI.pm and Writer directory in there but we'll save that for later. If you don't have the distribution handy you can navigate this at the BioPerl CVS browser.

Let's use the blast.pm module as an example to describe the relationship of the modules in this directory (could have substituted any of the other format parsers like fasta.pm or blastxml.pm - these are always lowercase for historical reasons). The module has some features you should look for - the first is the hash in the BEGIN block called %MAPPING. These key/value pairs here are the shorthand for how we map events from this module to general event names. This is only necessary because if we have an XML processor (see the blastxml.pm module) the event names will be the same as the XML tag names (like Hsp_bit-score in the NCBI BLAST XML DTD). So to make this general we'll make sure all of the events inside our parser map to the values in the %MAPPING hash - we can call them whatever we want inside this module. Some of the events map to hash references (like Statistics_db-len) and this is so we can map multiple values to the same top-level attribute field but we know they will be stored as a hash value in the subsequent object (in this example, keyed by the name dbentries). The capital "RESULT", "HSP", or "HIT" in the value name allow us to encode the event state in the event so we don't have to pass in two values. It is also easy for someone to quickly read the list of events and know which ones are related to Hits and which ones are related to HSPs. The listener in our architecture is the Template:PM. This object is attached as a listener through the Template:PM method add_EventListener. In fact you could have multiple event listeners and they could do different things. In our case we want to create Bio::Search objects, but an event listener could just as easily be writing data directly into a database or writing to a file, based on the events. The SearchResultEventBuilder takes the events thrown by the SearchIO classes and builds the appropriate Template:PM object from it.

Sometimes special objects are needed that are extensions beyond what the Template:PM or Template:PM objects are meant to represent. For this case we have implemented Template:PM so that it can use factories for creating its resulting Bio::Search objects - see the Template:PM::_initialize method for an example of how this can be set.

Writing and formatting output

Often people want to write back out a BLAST report for users who are most comfortable with that output or if you want to visualize the context of a weakly aligned region and use human intuition to score the confidence of a putative homologue. The Template:PM modules are for creating output using the information.

Template:PM currently creates output in a few different formats: text (recreating something like the BLAST report itself, in part or entirely), HTML, BSML, "ResultTable" (tab-delimited format), "HSPTable" (tab-delimited, for HSPs), and Gbrowse GFF.

The simplest way to output data in HTML format is as follows. <perl> my $writerhtml = new Bio::SearchIO::Writer::HTMLResultWriter(); my $outhtml = new Bio::SearchIO(-writer => $writerhtml,

                                -file   => ">searchio.html");
  1. get a result from Bio::SearchIO parsing or build it up in memory

$outhtml->write_result($result); </perl> If you wanted to get the output as a string rather than write it out to a file, simply use the following. <perl>


</perl> The Template:PM supports setting your own remote database url for the sequence links in the event you'd like to point to your own SRS or local HTTP-based connection to the sequence data. Simply use the remote_database_url method which accepts a sequence type as input (protein or nucleotide).

You can also override the id_parser() method to define what the unique IDs are from these sequence ids in the event you would like to use something other than the accession number that is gleaned from the sequence string.

If your data is instead stored in a database you could build the Bio::Search objects up in memory directly from your database and then use the Writer object to output the data.

Extending Template:PM

The framework for Template:PM is just a starting point for parsing these reports and creating objects which represent the information. If you would like to create your own set of objects which extend the current functionality we have built the system so that it will support this. For example, you may have built your own HSP object which supports a special operation like realign_with_sw(), which might realign the HSP via a Smith-Waterman algorithm, pulling extra bases from the flanking sequence. You might call your module Bio::Search::HSP::RealignHSP and put it in a file called Bio/Search/HSP/RealignHSP.pm. Note that you don't have to put this file directly in the BioPerl source directory - you can create your own local directory structure that is in parallel to the BioPerl release source code as long as you have updated your PERL5LIB to contain your local directory or you are using the "use lib" directive in your script. Also, you don't have to use the namespace Template:PM as namespaces don't mean anything to Perl with respect to object inheritance, but do we recommend you name things in a logical manner so that others might read and understand your code (and if you feel encouraged to donate your code to the project it might easily integrated with existing modules).

So, you're going to write your new special module, you do need to make sure it inherits from the base Template:PM object. Additionally unless you want to reimplement all the initialization state in the current Template:PM you should just plan to extend that object. You need to follow the chained constructor system that we have set up so that the arguments are properly processed. Here is a sample of what your code might look like (don't forget to write your own POD so that it will be documented, we've left it off here to keep things simple). <perl> package Bio::Search::HSP::RealignHSP; use strict; use Bio::Search::HSP::GenericHSP; use vars qw(@ISA); # for inheritance @ISA = qw(Bio::Search::HSP::GenericHSP); # RealignHSP inherits from GenericHSP

sub new {

 my ($class,@args) = @_;
 my $self = $class->SUPER::new(@args); # chained contructor
 # process the 1 additional argument this object supports
 my ($ownarg1) = $self->_rearrange([OWNARG1],@args);
 return $self; # remember to pass the object reference back out


sub realign_hsp {

 my ($self) = @_;
 # implement my special realign method here

} </perl> The above code gives you a skeleton of how to start to implement your object. To register it so that it is used when the Template:PM system makes HSPs you just need to call a couple of functions. The code below outlines them. <perl> use Bio::SearchIO; use Bio::Search::HSP::HSPFactory; use Bio::Search::Hit::HitFactory;

  1. setup the blast parser, you can do this with and SearchIO parser however

my $searchio = Bio::SearchIO->new(-file => $blastfile,

                                 -format =>'blast');
  1. build HSP factory with a certain type of HSPs to make
  2. the default is Bio::Search::HSP::GenericHSP

my $hspfact = Bio::Search::HSP::HSPFactory->new(-type =>

  1. if you wanted to replace the Hit factory you can do this as well
  2. additionally there is an analagous
  3. Bio::Search::Result::ResultFactory for setting custom Result objects

my $hitfact = Bio::Search::Hit::HitFactory->new(-type =>


$searchio->_eventHandler->register_factory('hsp', $hspfact); $searchio->_eventHandler->register_factory('hit', $hitfact); </perl> We have to register the HSPFactory, which is the object which will create HSPI objects, by allowing this to be built by a factory rather than a hard-coded Bio::Search::HSP::GenericHSP->new(...) call. We are allowing the user to take advantage of the whole parsing structure and the ability to slot their own object into the process rather than re-implementing very much. We think this is very powerful and worth the system overhead, but it may not permit this to be as efficient in parsing as we would like. Future work will hopefully address speed and memory issues with this parser. Volunteers and improvement code are always welcome.

Speed improvements with lightweight objects

The approaches described above will create a lot of objects, one for each of the components of a report. When you have 2000 hits in a BLASTX result there will be quite a few objects built, and a lot of memory consumed. It's possible that you'll want to use an approach that's less memory-intensive if your result sets are large. One option is to use the tabular output from BLAST when dealing with large datasets (-m 8 or -m 9).

There are other workarounds depending on what kind of data you want. We designed Template:PM to be a modular system which separates parsing the data from instantiating objects by throwing events (like SAX) and having a listener build objects from these events. So one can instantiate a different listener which builds simpler objects and throws away the data you don't want.

Here is an example of such a lightweight listener - Bio::SearchIO::FastHitEventBuilder - it just throws away the HSPs and only builds Result and Hit objects. <perl> use Bio::SearchIO; use Bio::SearchIO::FastHitEventBuilder; my $searchio = new Bio::SearchIO(-format => $format, -file => $file);

$searchio->attach_EventHandler(Bio::SearchIO::FastHitEventBuilder->new); while( my $r = $searchio->next_result ) {

 while( my $h = $r->next_hit ) {
  # Hits will NOT have HSPs
  print $h->significance,"\n";

} </perl> You could also build your own listener object - SearchResultEventBuilder and FastHitEventBuilder are two example implementations that specify the type of Result/Hit/HSP objects that are created by the listeners. You could create some lightweight Hit and HSP objects and have SearchResultEventBuilder create these instead of the default full-fledged ones.

The whole parser/listener design assumes that you want to process all the data for a result before moving on to the next one. From the listener's standpoint this means you have to store all the data you just got from the parser. Whether this is in memory, or potentially stored in a temporary file or database, would be up to the implementation.