[cmview-devel] Scoring functions

[Henning Stehr]

Hi Dan,

> If you implement this, I'll volunteer to plug in some of the
> residue-residue pairwise potentials from the literature. I'd really
> love to compare my scores to some of those described previously.

That would be great.

The nice thing about interfaces is that it's really up to you what you
implement on either side of the interface. So if we have several
scoring functions, they can be used in CMView as well as in a decoy scoring
suite using the interface. We just have to put some effort in the definition
because the whole purpose is of course that the interface won't change all
the time.

As you say, the scoring function could be based on any property, including
3D structure which then implies distance, atomic positions etc. The only issue
here is that at the interface, the scoring has to be residue based, i.e. the
function needs to return a score for a pair of residues (or a set of pairs).

Cheers,
Henning

On Wed, Apr 21, 2010 at 6:25 PM, Dan Bolser <dan.bolser at gmail.com> wrote:
> Great idea.
>
> If you implement this, I'll volunteer to plug in some of the
> residue-residue pairwise potentials from the literature. I'd really
> love to compare my scores to some of those described previously.
>
> Without the impediment of any knowledge of the existing architecture,
> I'd like to ask the following...
>
> * Can we (in theory) easily add scores based on arbitrary
> residue-residue interaction criteria?
>
> * How about scores based on explicit distance potentials? i.e. when
> the score is an explicit function of the distance of two particular
> groups?
>
> * Can we ... add scores based on reduced amino acid or atom alphabets?
>
> * Can we ... add scores based on NCI [1] for example?
>
>
> I'm starting to guess that the answer to all these questions is, "Yes,
> it's up to the method that supports the interface as to how the actual
> calculation is done". Is that correct?
>
>
> Thanks for starting this project, and thanks for creating open source software!
>
> All the best,
> Dan.
>
> [1] http://www.ncbi.nlm.nih.gov/pmc/articles/PMC168935/
>
> On 21 April 2010 16:48, Henning Stehr <stehr at molgen.mpg.de> wrote:
>> Matthias had the idea to define an interface for contact based scoring
>> functions such that new scoring functions can easily be added to
>> CMView. This would be very useful for ab-initio modelling and
>> refinement and to standardize things already implemented.
>>
>> Below you can see what I (in my role as architecture astronaut) think
>> such an interface ScoringFunction should look like.
>>
>> A scoring function could be based on the sequence (like a contact type
>> potential), contact map (like a contact count potential) or 3D
>> structure (like a geometric potential).
>>
>> In CMView, the score would be displayed as an overlay similar to the
>> way distance map or contact density are currently shown. One could
>> also add functions for refinement/editing based on the scores. Some
>> similar functionality is already implemented for DeltaRank scoring.
>>
>> For a start, it should be relatively easy to implement the scoring
>> functions described in:
>>
>> Dan M Bolser, Ioannis Filippis, Henning Stehr, Jose Duarte and Michael Lappe.
>> Residue contact-count potentials are as effective as residue-residue
>> contact-type potentials for ranking protein decoys (2008).
>> BMC Structural Biology 2008, 8:53: doi:10.1186/1472-6807-8-53
>>
>> In addition, we are currently working on two novel scoring functions
>> DeltaRank and GMBP which are still unpublished. It will still take
>> some time until they can be incorporated in such a way that they can
>> be used by external users.
>>
>> As said before, these are only ideas so far but it seems to make sense
>> to still include this in CMView 2.0.
>>
>> I'm posting this here for different reasons:
>> - as my online notebook so I won't forget
>> - to put this out for discussion and get feedback from other astronauts
>> - so that someone who feels the urge can start implementing something
>> - to promote the open source spirit of the project
>>
>> Happy hacking,
>> Henning
>>
>> --------------------
>>
>> interface ScoringFunction {
>>
>> /**
>> * Initializes the ScoringFunction object with the data that the
>> scoring can potentially be based on.
>> */
>> public void init(sequence, contacts, coordinates);
>>
>> /**
>> * Notifies the ScoringFunction object that the underlying data has
>> changed and scores have to be recalculated
>> */
>> public void updateData(sequence, contacts, coordinates);
>>
>> /**
>> * Returns the normalized score for contact (i,j). If (i,j) is not
>> really a contact, should
>> * returns the score assuming that it was a contact.
>> */
>> public double getScore(i,j);
>>
>> /**
>> * Returns a score for the whole contact map, as it would be used for
>> decoy scoring.
>> * This can be simply the sum of individual scores or something more complex.
>> */
>> public double getOverallScore();
>>
>> /**
>> * Returns a summed score for the current set of selected contacts.
>> Similar to above.
>> * Not sure if this is very useful but could be used to find good/bad regions.
>> */
>> public double getScoreForSelection(selected_contacts);
>>
>> }
>>
>> _______________________________________________
>> cmview-devel mailing list
>> cmview-devel at bioinformatics.org
>> http://www.bioinformatics.org/mailman/listinfo/cmview-devel
>>
>