[Joelib-help] Re: Convert help

[Joerg K. W. <we...@in...>]

Dear Fr=E9d=E9ric,

1. basic and slow way:
1.1. calculate all descriptors:
sh calculateDescriptors.sh +jcc inFile.sdf outFile.sdf

1.2. calculate descriptor statistic to check descriptor occurence.
sh statistic.sh -iSDF outFile.sdf
this will generate
outFile.sdf.statistic
outFile.sdf.binning

Because BCUT depends on the molecule size, large matrix entries, e.g.=20
Burden_modified_eigenvalues:Graph_potentials:18 (number=3Dinner=20
topological distance) occur not for small molecules.

You should only use BCUTS where every molecule has this value, otherwise=20
you should use a 'missing value'-replacing strategy, see also one of my=20
papers for more references:
http://www-ra.informatik.uni-tuebingen.de/publikationen/2004/wegner04jcic=
sA.html

Programs like Dragon hides this problems so they select only a specific=20
number of eigenvalues and uses often 0 for missing values, because the=20
eigenvalues are sorted in decreasing order. But you can not expect that=20
this will give really good models, if the descriptor contains a lot of=20
zeros.

1.3. select descriptors
sh descSelection.sh -iSDF outFile.sdf -oSDF selected.sdf descs.txt=20
normal " "

where descs.txt contains all descriptors your interested in in the form:
Burden_modified_eigenvalues:Graph_potentials:4
Burden_modified_eigenvalues:Graph_potentials:3
Burden_modified_eigenvalues:Graph_potentials:2
Burden_modified_eigenvalues:Graph_potentials:1
Burden_modified_eigenvalues:Graph_potentials:0
...

2. developer way:
joelib.test.DescriptorCalculation uses a helper class for this kind of=20
descriptors, which is joelib.desc.util.AtomPropertyDescriptors

if you use:
setDescriptors2Calculate(new String[]{"Burden_modified_eigenvalues"});
instead of the default settings you will get only the BCUT entries.

You can use this class on your own or you can modify the default-jcc=20
descriptor set in the variable defDescNames
in line -185- in DescriptorCalculation.

The arrays generated by the AtomPropertyDescriptors class are=20
automatically mapped to single value descriptors for which you can=20
calculate the statistic.

So, with some 'playing around' you should be able to store such results=20
directly in a database via JDBC, directly, by avoiding the overhead by=20
calculating all the other descriptors and the storing in files.

Kind regards, Joerg

> Dear Joerg,
> It's me again. Could you tell me the way to compute only BCUT descripto=
rs using the CalculateDescriptors.sh and more generally how can I select =
the descriptors for computation ?
> Thanks for your help
> Fred
>=20
> -----Original Message-----
> From: Joerg K. Wegner [mailto:we...@in...]=20
> Sent: Tuesday, August 24, 2004 12:08 PM
> To: JOELib help
> Cc: Fr=E9d=E9ric Ooms
> Subject: Re: Convert help
>=20
>=20
> Dear Fr=E9d=E9ric,
>=20
> beside the detailed bug report you can also try the extended descriptor=
=20
> calculation method provided by
>=20
> calculateDescriptors.sh -- help
>=20
> I recommend this one anyway. Why ?
> 1. The primitive calculation in convert calculates the descriptors by=20
> using default values (only one atom property).
> 2. The extended class
>     joelib.test.DescriptorCalculation
>     calculates the the descriptors for all available atom properties.
>     You can get a list by using:
>     sh convert.sh +sad
>=20
>     So this affects especially the BCUT, RDF and GTCI descriptor.
>     Furthermore for BCUT and RDF you have additional parameters, like t=
he
>     smoothing factor for RDF and the k-parameter in BCUT.
>     For RDF three different smoothing parameters are used, when using t=
he
>     extended calculation class.
>=20
>     Then these features can be rescaled using the one dimensional
>     Mahalanobis distance (z-score) normalization with mean=3D0 and
>     standardDeviation=3D1.
>=20
> Kind regards, Joerg
>=20
>=20
>>Dear Fr=E9d=E9ric,
>>
>>please add a bug request with detailed message, system, and example=20
>>file
>>to the tracking system:
>>
>>http://sourceforge.net/tracker/?group_id=3D39708&atid=3D425969
>>
>>Kind regards, Joerg
>>
>>
>>>Dear collegues,
>>>I a new JOELIB user as well as a Java newbiess... I am using the
>>>following command line to compute descriptors on a molecule
>>>Convert -h +d test.mol
>>>But I get the following error message:
>>>Exception in thread "main" java.lang.NullPointerException at=20
>>>joelib.Util.JOEPropertyHelper.getProperty....
>>>Could you help me ?
>>>Regards,
>>>Fred Ooms
>>>---------------------------------------------------------------
>>>Fr=E9d=E9ric Ooms, Ph.D.
>>>Chemistry Project Manager, Euroscreen S.A.
>>>Rue Adrienne Bolland, 47
>>>B-6041 Gosselies
>>>Tel. +32 71 348 500
>>>Fax: +32 71 348 519
>>>
>>>
>>
>>
>=20
>=20


--=20
Dipl. Chem. Joerg K. Wegner
Center of Bioinformatics Tuebingen (ZBIT)
Department of Computer Architecture
Univ. Tuebingen, Sand 1, D-72076 Tuebingen, Germany
Phone: (+49/0) 7071 29 78970
Fax: (+49/0) 7071 29 5091
E-Mail: mailto:we...@in...
WWW:    http://www-ra.informatik.uni-tuebingen.de
--
Never mistake motion for action.
                                     (E. Hemingway)

Never mistake action for meaningful action.
                                (Hugo Kubinyi,2004)