Molecular descriptor sets

Molecular descriptors are powerful tools for the approximation of selected properties of chemical structures in an easy-to-handle form that allows efficient comparison and selection of compounds possessing required chemical, structural, pharmacological or biological features. These methods are not only faster than structure based approaches, where chemical structures or other high level structural models are compared against each other directly, but the accuracy of descriptor based approaches in estimating properties can be as good as the accuracy of structure based approaches.
One disadvantage of the use of descriptors, however, is that one descriptor can capture certain but apparently not an unlimited number of properties of the chemical structure encoded.

The concept of molecular descriptor set has been introduced in order to overcome this deficiency. A molecular descriptor set combines an arbitrary number and type of molecular descriptors into one single entity. Each component aim to depict a certain characteristic of the original compound and their combination is very information rich thus descriptor sets are even more powerful in screening large compound libraries for complex criteria than single descriptors are.

Generating descriptors

Molecular descriptors are generated from molecular structures. Although different descriptors utilize different processing steps, still there are numerous steps common to these procedures. The building up of the molecular structure itself - which is an unavoidable initial step of all kinds of descriptor generation - provides an apparent example: it has to be loaded from either a molecular file or from a database, then appropriate sized memory structures have to allocated and initialized according to the data read. Following these costly steps the molecule has to be standardized. Standardization is another tedious process that may include aromatic ring perception, counter ion removal and various further transformations of the initial structure. The efficiency of descriptor generation can be improved if several descriptors are generated simultaneously, as the above outlined common steps are not repeated superfluously.

The command line program GenerateMD supports the generation of one descriptor in one go. This means, that multiple descriptors belonging to the same molecular structures can be generated consecutively by running GenerateMD several times. Although this is more time-consuming but much simpler to use. In this case the ease of use was deliberately chosen over efficiency. Note, however, that the Application Program Interface allows the generation of multiple descriptors at the same time.

Storing descriptors in database

Descriptors belonging to a molecule, can be stored in relational databases. There is no limitation on the number of descriptors associated to a structure, but there is no predefined table structure which is automatically maintained by the program. Different descriptors are stored in different tables. Users are not required to be aware of details of all these. All they need to remember is the name of the structure table where molecules originated from and the name of the descriptor, given by the user itself at the time of descriptor generation.

The type of database table columns storing descriptors depends on the type of descriptor and the database management engine used. The table below summarizes supported RDBMS-s and implemented descriptors indicating the corresponding column type:

Table 1 Column types of various molecular descriptors in different Relational Database Management Systems.

Commands

Commands:
   c       creates new descriptor, default
   u       updates existing descriptor in database
   d       deletes existing descriptor from database
   a       adds new screening configurations to existing descriptors
   r       removes screening configurations
   l       list all screening configurations

Command syntax:
   c [input file] [options]
   u <structure table name> <descriptor name>
   d <structure table name> <descriptor name>
   a <structure table name> <descriptor name> [ <config. name>] <config. file>
   r <structure table name> <descriptor name> <configuration name> ...
   l [ <structure table name> ]

The most typical use of GenerateMD is the generation of molecular descriptors. However, some other operations that are closely related to the generation of descriptors. These arise only in the case when molecular descriptors are stored in a database. Such descriptors can be manipulated with GenerateMD, moreover, it is highly recommended to avoid using general purpose database management tools since the integrity of structure tables and descriptors can easily be violated.

Descriptors generated earlier needs to be updated when new structures are imported into the structure table which the molecular descriptors are associated with. This can be done with the u (update) command.

Deleting old, unwanted descriptors should be done by the d commands (rather than issuing an SQL DROP TABLE statement).

Along with the descriptors generated, some other information are stored in the database: the name of the descriptor (for example pharma-for-H1), the type of the descriptor (for example PF), and the configuration used for the descriptor generation. The user of the program need to remember only the name (since it was given by him or her), although he or she should be aware of the configuration stored. Whenever the descriptors are in use (for instance used in virtual screening), this original configuration is applied. However, this may not always be sufficient, for instance new metrics might be needed for screening. Three more commands are provided to extend this original configuration.
The command a (add) allows the extension of the original configuration. See example how to add a new metric (optimized for another active family, but using the already generated descriptor). The configuration name given as the third parameter in the command line is stored in the database, and is used to refer to the particular screening configuration. When this name is omitted, the configuration name is taken from the configuration file name (by removing the file path and extension).

Command l can be applied to list existing screening configurations, and r to remove any previously added configurations.

Options

Options and parameters can either be defined in the command line or be specified in an XML configuration file. The command line mode is more suitable for smaller experiments. In contrast to this, configuring generatemd from XML is more user-friendly and convenient. However, it is only the creation of new descriptors that is available through XML configuration. In the case of the other command it would be unnecessarily time-consuming to create the configuration files instead of passing a few parameters in the command-line. There is an example configuration file available.

General options:
  -h, --help               this help message
  -x, --expert-help        advanced options for expert users
  -L, --list-descriptors   lists all available built-in descriptor types
  -v, --verbose            verbose
  -s, --saveconf           saves database settings

Input/output options:
  -a, --table-name <table> takes structures from the given database table
  -q, --query <sql>        where-clause for reading structures from a table
  -o, --output <filepath>  SDF output file path and name
  -g, --generate-id [<first>]
                             generate unique structure identifiers
                             an optional value for the first ID can be given
  -S, --sdf-output         generate SDfile output (otherwise descriptor file)

Database options:
  -d, --driver <JDBC>      JDBC driver
  -u, --dburl <url>        URL of database
  -l, --login <login>      login name
  -p, --password <pwd>     password

Descriptor options:
  -k, --descriptor <type> <name> [ <comment> ]
                           generates descriptors of the given type and name
                             in database
  -k, --descriptor <type>
                           generates descriptors of the given type in file
  -c, --config <config>    path and name of the XML configuration file


Input/output options:
  -D, --decimal-format     tab separated output of the descriptor values
  -2, --binary-format      0 1 string
  -T, --stat               print statistics on descriptors generated

SDfile options:
  -I, --id-tag <name>      name of the tag storing unique molecule identifiers
  -t, --use-tag <name>     use the named tag instead of the default one to
                             store descriptor data
  -P, --PMAP-tag [<name>]  use existing PMAP data
  -w, --validate <name>    validate the descriptor using activity-seeded,
                             structure-based clustering, <name> is an SDfile
                             tag containing the IC50 activity data

Database options:
  -O, --proptable [<table>]  name of the property table
  -U, --update-on-insert   update descriptors when new structure is inserted

Chemical fingerprint options:
  -f, --fingerprint-length chemical fingerprint length in bits
  -n, --pattern-length     maximum number of bonds in patterns
  -b, --bit-count          maximum number of bits to set for each patterns

Fuzzy smoothing options:
  -z, --fuzzy <smoothing factor>
                           generate fuzzy pharmacophore fingerprints with the
                             given fuzzy smoothing factor
  -G, --Gaussian-cutoff <value>
                           not smoothing outside <value>*sigma
  -B, --smoothing-bound <value>
                           lower bound for fuzzy smoothing
  -A, --asymmetrical-smoothing
                           do not smooth left side
  -R, --ignore-rotomers    do not take into account the effect of rotatable
                             bonds

Merging short forms of command line options is not supported (that is, instead of -Tv the form -T -v should be used).

In the XML configuration file the same parameters as in the command line can be defined. The command line options appear in the config configuration editor labeled with the above long forms of command line parameter names, with only small differences. In most of the cases only the first letters of words are capitalized. For example --asymmetrical-smoothing is displayed as AsymmetricalSmoothing.

Input

The software may take molecules from either a text file or from a database. The input file can be specified following the command type. If the file name is omitted, the standard input is read, except when the option -q, --query is defined in which case input is taken from a database. Input files should be molecular structure files of one of the common formats like MDL molfile, Compressed molfile, SDfile, Compressed SDfile, SMILES (see the Administration Guide of JChem for a full list of molecular structure files accepted).

If the input file is an SDfile, it may already contain the descriptor (or some of them) to be generated, or other data needed for descriptor generation (for instance, the pharmacophore map (the list of pharmacophore points) which is required for pharmacophore fingerprint generation). Such existing data can either be used or omitted during descriptor generation. The default behavior of GenerateMD is to omit such information. In order to use it the -P, --PMAP-tag flag has to be specified. The default SDfile tag for the pharmacophore map is PMAP. Other than default tag name can be specified after the -P, --PMAP-tag option.

Output

GenerateMD is capable of writing result descriptor sets either into a text file (-o, --output) or into a database table when database options are defined (either explicitly by the appropriate flags or implicitly by a JChem user configuration file .jchem in the user's home directory). If no output is specified, results are written to standard output as readable text.

Descriptors

When creating or updating descriptors, one -k flag is mandatory. Descriptor type specific flags always follow the this flag. GenerateMD supports the following descriptor types: CF, PF, Burden eigenvalue descriptor (or BCUT) and various scalar descriptors. Besides, the generation of user-defined custom descriptors is also supported.

File output

In the case of SDfile input an SDF output can be generated by the use of the -S flag. Otherwise, a descriptor file is created. Such descriptor files are readable text files that can be processed by other applications, e.g. by ScreenMD. Descriptor files contain one descriptor per line along with a unique identifier. The first two lines form the file header that should be ignored when the file is read by users.

Descriptors generated by GenerateMD can be used in JKlustor, too. To produce descriptors in a format accepted by JKlustor tools, the decimal output format options has to be specified with the -D flag. Using -D does not imply -g, i.e. unique identifiers are not generated, and -g has to be set explicitely if required. Besides the automatically generated identifiers, natural identifiers given in the input SDfile can also be printed in the JKlustor typed output file. For this purpose the flag -I, --id-tag can be used.

Database output

Descriptors generated can be stored in database tables. The appropriate table is created and maintained automatically. The user of the program should only be aware of the name of the descriptor given at the time of generating the descriptor (second parameter after the -k flag). This name is used to access the descriptor data from other applications. Descriptors can either be inserted into empty rows (command c) or alternatively existing data can be updated (command u).

An important database related option is -U, --update-on-insert. Descriptors created with this option selected will be updated (i.e. generated) whenever a new structure is inserted into the corresponding structure table. If this option is not specified in the command line, the descriptor table is not updated automatically, however, at each use (e.g. similarity searching) a warning message is given that the regeneration of the descriptors is required because the structure table contains newer structures.

Statistics

Simple statistics on the generated descriptors can also be obtained as an output of GenerateMD by specifying the -T (--stat) option. This is always written to the standard output. It serves as a first-hand aid to evaluate and tune descriptor parameters.

Configuration Files

Besides the XML configuration file that can be optionally used to specify parameter settings, the GenerateMD application may take other configuration files too. Such files correspond to the molecular descriptors to be generated.

Different descriptor types require different parametrization. The actual parameter settings are defined in external text (XML) files. One such file corresponds to one descriptor type.