Molecules from the Minkowski space: an approach to building 3D molecular structures
In the field of computational chemistry it is usual to have only a partial set of structural information about compounds, like the connectivity or the formula. Individual studies can easily be performed using ‘human interfaces’ for building input structures. However, automatic, ‘batch’ processes cannot be applied on a large number of molecules if they imply human intervention. Studies, like QSAR, pharamacophore analysis, reaction prediction might need full, complete 3D information for the compounds of interest. The widespread tools used for structure determination (force-fields or quantum chemical methods) even require a complete set of initial 3D coordinates. Our approach intends on generating globally valid set of 3D coordinates for small and medium sized molecules, based on local structural criteria. Over against iterative, backtrack based structure predicting algorithms, our method is capable of satisfying partially inconsistent requirements. Such situations are common for structures holding polycyclic, rigid details. Goals mentioned above can be achieved using coordinates interpreted in a space with a Minkowski metric. Our coordinate assignment process is divided into the following parts: (I) Automatic generation of distance criteria based on chemically relevant local properties, such as bond stretches, bond angles, dihedral angles, etc. (II) Multi-dimensional coordinate assignment which fulfills all the criteria. (III) Geometry optimization using a force field extended to the multidimensional Minkowski space. The optimization eliminates the over-3D components and yields the 3D coordinates.