[Biojava-dev] Pairwise Alignment methods

Fri Jan 25 04:25:23 UTC 2008

Hello again :-)

On Jan 25, 2008 1:43 AM, Mark Schreiber <markjschreiber at gmail.com> wrote:

> On Jan 25, 2008 10:06 AM, Felipe Albrecht <felipe.albrecht at gmail.com>
> wrote:
> > Hi,
> >
> > is not possible to add into the SequenceAlignment interface something
> like:
> > "double doAlignmentAndGetTheScore(SymbolList symbolList1, SymbolList
> > symbolList2)"?
> > Okay, the name is horrible, but you know what it means.
> >
>
> You could but you would break backwards compatibility with anyone who
> implemented this interface previously.  We have sometimes done this in
> biojava but we would need to make sure it will break no ones code.
> Another option would be to extend the interface with another that adds
> this method (not very tidy I know).

If I'm correct, the SequenceAlignment is an abstract class, so, we can
define there with an empty implementation, and SmithWaterman and others
classes implements it. Anyone that implemented SequenceAlignment will not
see anything different.

>
> >
> > > While ultra optimized algorithms where once a major
> > > feature of bioinformatics this is becoming less necessary as standard
> > > desktops are now equivalent to the super computers of 5 years ago.
> >
> > Okay, but do not forget that the bioinformatics data size is growing
> faster
> > than the computer processing and main memory capacity.
> >
> > What im trying to say is that the actual methods are fast [and light]
> enough
> > for do 1, 10, 100, 1000 pairwise alignments, but not for 10k, 100k or in
> my
> > case, 65k * 65k.
>
> One could also argue that Smith Waterman is not ideal for large
> sequences. I think it is o(NM) or something.

I'm not comparing  two sequences with 65k * 65k  bases, but doing the
alignment of 65k little sequences between then.

>
>
> > Really, I dont see problems of having optimized functions for specifics
> > operations, as unix phylosofies: "do small programs for specifics
> things,
> > for big things join then" (Something like it :-) ).
> >
>
> Yes, this would be an argument for workflow or service oriented
> architecture built from multiple inter operable biojava sub-projects.
> This is obviously not what biojava is. Indeed biojava is not even an
> application you just use it to build applications. Maybe for your use
> case you could use biojava to handle the I/O and then do the more
> efficient SW using your own code. BioJava is a collection of objects
> that are (somewhat) related and interoperable. It doesn't mean you
> have to use biojava throughout your application.

Okay, now I understood, biojava is not a library for bioinformatics
applications, but for interconnect bioinformatics applications. So, biojava
in the actual way is not appropriate for the application that I am
developing. I will develop some "optimized" classes and functions for my use
and when it will be ready I will announce in this mailing list and ask if
want to merge in biojava. If biojava team needs somebody to improve some
biojava functions, specially sequences and sequences IO, can ask me.

Thank you

Felipe Albrecht

>
>
> - Mark
>
> > On Jan 24, 2008 11:26 PM, Mark Schreiber <markjschreiber at gmail.com>
> wrote:
> > > Hi Felipe -
> > >
> > > I agree your method is more efficient but I think it violates the
> > > SequenceAlignment interface which would cause compatibility problems.
> > > I also wonder what should happen if a user calls the getAlignment()
> > > method if you have only calculated a score.
> > >
> > > instanceof is potentially expensive but it is nothing compared to
> > > actually performing the SmithWaterman.
> > >
> > > Biojava is somewhat memory heavy but this is largely because it is
> > > object oriented. Certainly something in C would be lighter and faster
> > > but the whole point in using Java is the relative benefits of object
> > > oriented design.  While ultra optimized algorithms where once a major
> > > feature of bioinformatics this is becoming less necessary as standard
> > > desktops are now equivalent to the super computers of 5 years ago.
> > >
> > > I actually find the SW and NW to be reasonably fast. This is because
> > > all the heavy lifting is done in loops that the JVM presumably
> > > compiles and executes natively.
> > >
> > > - Mark
> > >
> > >
> > >
> > >
> > >
> > > On Jan 25, 2008 3:40 AM, Felipe Albrecht <felipe.albrecht at gmail.com >
> > wrote:
> > > > Hello,
> > > >
> > > > I saw the commit and I think that this solution is not the better.
> > > > I think it because you are creating internally two Sequence and
> probably
> > the
> > > > programmer will not use others alignment information,  he will use
> only
> > the
> > > > score.
> > > >
> > > > Because it, I think that if you have 2 SymbolList, just do the
> alignment
> > and
> > > > return the score, as I did.Otherwise, If the programmer want the
> "visual
> > > > alignment", he should create externally the SimpleSequences, it is,
> not
> > the
> > > > method must do it.
> > > >
> > > > IMHO, one [serious] problem in biojava is the memory consumption, it
>
> > have
> > > > not "lightweight" classes or methods that do the things quickly.
> Because
> > it,
> > > > may be is a good choice to have a method that simply gives the
> alignment
> > > > score, and not do the others things, like backtracking. Another
> think,
> > the
> > > > cost of the "instanceof" is high.
> > > >
> > > > Thank you,
> > > >
> > > > Felipe Albrecht
> > > >
> > > >
> > > >
> > > >
> > > > On Jan 24, 2008 11:35 AM, Mark Schreiber <markjschreiber at gmail.com >
> > wrote:
> > > > > Hi -
> > > > >
> > > > > I have just commited changes that let you use SymbolLists in all
> parts
> > > > > of the NW and SW SequenceAlignment objects.
> > > > >
> > > > > As you suggested I made the matrix a method local variable. I also
> > > > > removed calls to the garbage collector.
> > > > >
> > > > > This can be checked out from SVN.
> > > > >
> > > > > - Mark
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > On Jan 24, 2008 9:05 PM, Felipe Albrecht <felipe.albrecht at gmail.com>
> > > > wrote:
> > > > > > If you prefer, I can send a diff and should I do the same thing
> in
> > > > > > SequenceAlignment and NeedlemanWunsch classes?
> > > > > >
> > > > > > Thank  you,
> > > > > >
> > > > > > Felipe Albrecht
> > > > > >
> > > > > >
> > > > > >
> > > > > > On Jan 24, 2008 5:50 AM, Mark Schreiber <
> markjschreiber at gmail.com >
> > > > wrote:
> > > > > > > Hi Felipe -
> > > > > > >
> > > > > > > Thanks for the input on this. As a general rule the GC should
> > never be
> > > > > > > called from code. Generally this degrades performance of the
> JVM.
> > > > > > > Unless there is a very good reason I will remove this.
> Probably
> > you
> > > > > > > are right a method parameter may work better.
> > > > > > >
> > > > > > > - Mark
> > > > > > >
> > > > > > > On Jan 24, 2008 1:47 PM, Felipe Albrecht
> > <felipe.albrecht at gmail.com >
> > > > > > wrote:
> > > > > > > > Hello,
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > >
> > > > > > > > I think that it can be solved by a simple way:
> > > > > > > > Implement (or just copy and cut) a pairwiseAlignment
> utilizing
> > > > SymboList
> > > > > > as
> > > > > > > > parameters and do no creating a alignment, just the
> calculating
> > it
> > > > and
> > > > > > > > returning the value.
> > > > > > > >
> > > > > > > > Another thing that is a bit stange for me, is the
> utilization of
> > > > garbage
> > > > > > > > collector direcly, that is: The field "scoreMatrix" is a
> class
> > > > field,
> > > > > > why at
> > > > > > > > the end of pairwiseAlignment it is set to null and the
> garbage
> > > > collector
> > > > > > > > run? It is not better (and simpler) to use scoreMatrix as
> method
> > > > > > variable?
> > > > > > > >
> > > > > > > > I'm annexing the class code with my changes that is doing
> well
> > the
> > > > (4^8)
> > > > > > *
> > > > > > > > (4^8) SymbolList pairwise alignments that I am needing :-)
> > > > > > > >
> > > > > > > > Thank you,
> > > > > > > >
> > > > > > > > Felipe Albrecht
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > > >  On Jan 23, 2008 6:50 AM, Mark Schreiber <
> > markjschreiber at gmail.com
> > > > >
> > > > > > wrote:
> > > > > > > > > Hi Felipe -
> > > > > > > > >
> > > > > > > > > I agree this is a barrier to ease of use. Even if
> Sequences
> > are
> > > > > > > > > required internally for some obscure reason there is no
> reason
> > why
> > > > > > > > > dummy Sequences cannot be made inside the aligner.  These
> > > > sequences
> > > > > > > > > could be given names like 'query' and 'subject' or even
> 'seq1'
> > and
> > > > > > > > > 'seq2'.
> > > > > > > > >
> > > > > > > > > I will take a look at adding some methods.
> > > > > > > > >
> > > > > > > > > Best regards,
> > > > > > > > >
> > > > > > > > > - Mark
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > >
> > > > > > > > > On Jan 23, 2008 2:58 PM, Felipe Albrecht
> > > > < felipe.albrecht at gmail.com >
> > > > > > > > wrote:
> > > > > > > > > > Hello all,
> > > > > > > > > >
> > > > > > > > > > I have a simple question about pairwise alignment
> classes
> > > > > > (SmithWaterman
> > > > > > > > and
> > > > > > > > > > NeedlemanWunsch):
> > > > > > > > > > Why it is necessary two Sequence for alignment and not
> two
> > > > > > SymbolList?
> > > > > > > > > >
> > > > > > > > > > Example, I have a SymbolList collection to align between
> > then,
> > > > > > > > > > by this way I need to create some "dummies"  Sequence
> for to
> > do
> > > > the
> > > > > > > > > > alignment.
> > > > > > > > > >
> > > > > > > > > > Reading the source, I saw that the unique field that is
> > > > exclusive to
> > > > > > > > > > Sequence is the name, for the alignment output,
> > > > > > > > > > but if I need only the alignment result, it is useless.
> > > > > > > > > >
> > > > > > > > > > It is not possible to override the pairwiseAlignment to
> > accept
> > > > > > > > SymbolList or
> > > > > > > > > > may be a new method that the parameters are 2 SymbolList
> and
> > > > returns
> > > > > > the
> > > > > > > > > > alignment score?
> > > > > > > > > >
> > > > > > > > > > Thank you
> > > > > > > > > >
> > > > > > > > > > Felipe Albrecht
> > > > > > > > > > _______________________________________________
> > > > > > > > > > biojava-dev mailing list
> > > > > > > > > > biojava-dev at lists.open-bio.org
> > > > > > > > > > http://lists.open-bio.org/mailman/listinfo/biojava-dev
> > > > > > > > > >
> > > > > > > > >
> > > > > > > >
> > > > > > > >
> > > > > > >
> > > > > >
> > > > > >
> > > > >
> > > >
> > > >
> > >
> >
> >
>