FastSearch Class Reference

#include <openbabel/fingerprint.h>

List of all members.

Public Member Functions
std::string	ReadIndexFile (std::string IndexFilename)
std::string	ReadIndex (std::istream *pIndexstream)
virtual	~FastSearch ()
bool	Find (OBBase *pOb, std::vector< unsigned int > &SeekPositions, unsigned int MaxCandidates)
bool	FindMatch (OBBase *pOb, std::vector< unsigned int > &SeekPositions, unsigned int MaxCandidates)
bool	FindSimilar (OBBase *pOb, std::multimap< double, unsigned int > &SeekposMap, double MinTani)
bool	FindSimilar (OBBase *pOb, std::multimap< double, unsigned int > &SeekposMap, int nCandidates=0)
OBFingerprint *	GetFingerprint () const
const FptIndexHeader &	GetIndexHeader () const

Detailed Description

The FastSearch class searches an index to a datafile containing a list of molecules (or other objects) prepared by FastSearchIndexer.

OpenBabel can also search files for molecules containing a substructure specified by a SMARTS string, using OBSmartsPattern or from the command line:

    babel datafile.xxx -outfile.yyy -sSMARTS

Note that the index files are not portable. They need to be prepared on the computer that will access them.

Using FastSearch and FastSearchIndexer in a program

The index has two tables:

• an array of fingerprints of the molecules
• an array of the seek positions in the datasource file of all the molecules

To prepare an fastsearch index file:

• Open an ostream to the index file.
• Make a FastSearchIndexer object on the heap or the stack, passing in as parameters:
• the datafile name, the indexfile stream,
• the id of the fingerprint type to be used,
• the number of bits the fingerprint is to be folded down to, If it is to be left unfolded, set fpid to 0 or do not specify it.

• For each molecule, call Add() with its pointer and its position in the datafile.
  Currently the std::streampos value is implicitly cast to unsigned int so that for 32bit machines the datafile has to be no longer than about 2Gbyte.
• The index file is written when the FastSearchIndexer object is deleted or goes out of scope.

To search in a fastsearch index file

• Open a std::istream to the indexfile (in binary mode on some systems)
• Make a FastSearch object, read the index and open the datafile whose name it provides

      ifstream ifs(indexname,ios::binary);
      FastSearch fs;
      string datafilename = fs.ReadIndex(&ifs);
      if(datafilename.empty()
         return false;
  
      ifstream datastream(datafilename);
      if(!datastream)
         return false;
  

To do a search for molecules which have all the substructure bits the OBMol object, patternMol

    vector<unsigned int>& SeekPositions;
    if(!fs.Find(patternMol, SeekPositions, MaxCandidates))
            for(itr=SeekPositions.begin();itr!=SeekPositions.end();++itr)
      {
         datastream.seekg(*itr);
         ... read the candidate molecule
         and subject to more rigorous test if necessary
      }

To do a similarity search based on the Tanimoto coefficient This is defined as:
Number of bits set in (patternFP & targetFP)/Number of bits in (patternFP | targetFP)
where the boolean operations between the fingerprints are bitwise
The Tanimoto coefficient has no absolute meaning and depends on the design of the fingerprint.

    multimap<double, unsigned int> SeekposMap;
    // to find n best molecules
    fs.FindSimilar(&patternMol, SeekposMap, n);
    ...or
    // to find molecules with Tanimoto coefficient > MinTani
    fs.FindSimilar(&patternMol, SeekposMap, MinTani);

    multimap<double, unsigned int>::reverse_iterator itr;
    for(itr=SeekposMap.rbegin();itr!=SeekposMap.rend();++itr)
    {
       datastream.seekg(itr->second);
       // ... read the candidate molecule
       double tani = itr->first;
    }

The FastSearchFormat class facilitates the use of these routine from the command line or other front end program. For instance:

Prepare an index:

    babel datafile.xxx index.fs

    babel datafile.xxx -ofs namedindex

• which type of fingerprint to be used, e.g. -xfFP2,
• whether it is folded to a specified number of bits, e.g. -xn128 (which should be a power of2)
• whether to pre-select the molecules which are indexed:
• by structure e.g only ethers and esters, -sCOC
• by excluding molecules with bezene rings, -vc1ccccc1
• by position in the datafile e.g. only mols 10 to 90, -f10 -l90

    babel index.fs outfile.yyy -sSMILES

    babel datafile.xxx outfile.yyy -sSMILES -ifs

With the -S option, the target can be specified as a molecule in a file of any format

    babel datafile.xxx outfile.yyy -Spattern_mol.zzz -ifs

Similarity search in a previously prepared index file
This rather done (rather than a substructure search) if the -at option is used,

    babel datafile.xxx outfile.yyy -sSMILES -ifs -at0.7

• -at0.7 will recover all molecules with Tanimoto greater than 0.7
• -at15 (no decimal point) will recover the 15 molecules with largest coefficients.
• -aa will add the Tanimoto coefficient to the titles of the output molecules.

All stages, the indexing, the interpretation of the SMILES string in the -s option, the file in the -S option and the final SMARTS filter convert to OBMol and apply ConvertDativeBonds(). This ensures thatforms such as[N+]([O-])=O and N(=O)=O are recognized as chemically identical.

Constructor & Destructor Documentation

Member Function Documentation

• fingerprint.h
• fingerprint.cpp