smi23d - 3D Coordinate generation

smi23d consists of two programs that can be used to convert one or more SMILES strings to 3D. The first step uses a program called smisdf which generates a set of rough 3D coordinates using an iterative refinement procedure. These coordinates are then optimized with a MMFF94 force field by mengine to generate a reasonable 3D structure. In addition to generating an optimized structure, the code can optinally calculate some molecular properties such as XlogP, dipole moment and vibrational properties. By default these are not calculated. The resultant coordinates are written in SD format.

Important Features

It should be noted that the programs do not aim to identify the lowest energy (or a low energy) conformer. The output of this procedure is simply a reasonable 3D structure for a given molecule, which can be used as a starting point for other investigations.
If you're molecules are polycyclic aromatics, it is a good idea to provide kekulized SMILES, rather than aromatic SMILES.

We have used these programs to convert 17M structures from Pubchem to 3D. These structures are available in the Pub3D database. In addition we have provided a web form where you can paste a set of SMILES and get back a set of 3D coordinates in SD format.

Updates

06/13/08 - Fixed a bug in SMILES parsing
04/16/08 - Fixed a bug in ring perception
03/24/08 - Updated code to correctly handle aromaticity detection when aromatic SMILES are specified. Also corrects some issues with long molecule names in SMI files.
02/29/08 - Cleaned up build system to properly install binaries and parameter files in the build directory. Also updated error handling code for missing input and parameter files for smi23d
02/15/08 - Updated force constant for N-Br bonds
02/06/08 - Updated mengine to build on Cygwin

Downloads

The source code for the programs can be obtained from the CICC-Grid Sourceforge SVN repository by doing

svn co https://cicc-grid.svn.sourceforge.net/svnroot/cicc-grid/cicc-grid/smi23d/trunk smi23d

If you'd rather not download and compile the code, we've made a few binary builds available. Note that you'll also need to download the parameters files: mmxconst.prm and mmff94.prm.

Platform
Intel OS X	smi2sdf	mengine
Linux, Fedora Core 5 Intel 64 bit	smi2sdf	mengine

Tobias Kind has kindly provided a version for Windows (compiled using Cygwin)

Compiling

The code is written in C and should not require any library beyond the standard C library. The build system uses Scons, a Python alternative to Makefile's. The sources are available from the CICC-Grid Sourceforge SVN repository. You can check out the smi23d trunk by doing the following

svn co https://cicc-grid.svn.sourceforge.net/svnroot/cicc-grid/cicc-grid/smi23d/trunk smi23d

This will create a directory called smi23d. Change into the directory and then type

scons

This will compile the code and place the binary executables along with the parameter files into the build directory. Note that the binaries are built with no optimization. It has been observed that using -O2 on 32 bit Linux leads to binaries that give the wrong results.

However using the -ffast-math option to gcc, does lead to significant speedups. Tobias Kind also reports a set of gcc parameters that leads to a 5x speedup for some test cases, though as above, it employs the -O3 optimization level, so in some cases, might give non-reproducible errors.

Usage

In general one would start from a set of SMILES strings which are to be converted to an SD file containing 3D coordinates. The first step is to generate a set of rough 3D coordinates

smi2sdf -o rough.sdf -p mmxconst.prm mymols.smi

You can provide the name of any output file. Also you have to give it the full path to the parameter file. If not specified it will look in the current working directory. The program will generate a number of extra files including an error log (error.log) and a SMILES containing the entries from the input file that could not be processed due being inorganic.

Once you have generated a set of rough coordinates, we use this file (in our example rough.sdf) as input to mengine to get the optimized 3D coordinates.

mengine -p mmff94.prm -c mmxconst.prm -o opt.sdf rough.sdf

This will dump output to the STDOUT and place the optimized coordinates into opt.sdf. mengine can evaluate a number of molecular properties include dipole moment, XlogP and various vibrational properties. By default these are not calculated. Run mengine with no arguments to get the help page which describes how to evaluate each of these properties.

In case you don't want to download and compile the program we provide a SOAP web service that can be used to convert SMILES strings to SDF (returned as a string). You can get the WSDL and documentation describing the arguments that the service requires. We have also made available a web page client (limited to 200 molecules at a time).

Benchmarks

Since the programs do not aim to generate a low energy conformer, comparison with other programs can be difficult. Since we do not have access to other coordinate generators, we welcome anybody who can perform such comparisons. We have done some timing benchmarks using molecules from PubChem ranging in size from 5 to 100 heavy atoms. You can view plots of time versus heavy atom count for smi2sdf and mengine