Automatic 3D Structure Generation

The program system COBRA [118, 119] can be regarded as a rule- and data-based approach, but also applies the principles of fragment-based (or template-based) methods extensively (for a detailed description sec Chapter 11, Sections 7.1 and 7.2 in the Handbook). COBRA uses a library of predefined, optimized 3D molecular fragments which have been derived from crystal structures and foi ce-field calculations. Each fi agment contains some additional information on 

In the next step, the suggested models are translated into 3D space by subsequently combining the templates. Again, each model is assessed and ranked according to various structural criteria, such as the geometric fit of the 3D templates and non-bonding interactions (steric clashes). If none of the solu- 

In the next step, the molecule is fragmented into ring systems and acyclic parts. Ring systems are then separated into small and medium-sized rings with up to nine atoms and into large and flexible systems. 

The number of possible conformations rises dramatically with increasing ring size. Therefore, large ring systems carmot be handled by the methods applied to small rings. However, in rigid polymacrocyclic structures for example, an overall 

After generating the entire ring system of the molecule, CORINA uses a reduced force field to optimise the ring geometries. Two simplifications lead to this so- 

---

This section describes briefly some of the basic concepts and methods of automatic 3D model builders. However, interested readers are referred to Chapter II, Section 7.1 in the Handbook, where a more detailed description of the approaches to automatic 3D structure generation and the developed program systems is given. 

Here, an attempt to classify different strategies to generate 3D molecular models is undertaken with the aim to specify the remit of methods which will be covered under the term automatic 3D structure generators . The focus will be on methods designed for small, dmg-like molecules. The prediction of the geometry of polymers, in parhcular of biopolymers, is a task of its own and not even attempted by the approaches discussed here. 

Conformation analysis methods. In many cases in the process of building a 3D structure from scratch, decisions have to be made between multiple alternatives with similar energy. A typical example is an sp -sp torsion angle with similar energies for the alternatives of -i-60°, -60° and 180°. In many cases, rules are used to decide (e.g. stretch an open chain portion as much as possible to avoid clashes). Sometimes, the best result cannot be determined without a conformation analysis (e.g. complex ring systems with exocycHc substituents). Despite conformation analysis being a topic of its own covered in the next chapter, many automatic 3D structure generators have to fall back in certain situations to a limited conformation search in order so solve a specific problem and to come up with a reasonable solution. 

Distinct from these are automatic methods that directly transform 2D input information on atoms, bonds, and stereochemistry into 3D atomic coordinates. The automatic methods are classified into rule-based and data-based, fragment-based, conformational analysis, and numerical methods (Fig. 5). These classes of methods overlap more or less with each other and belong more or less to the domain of automatic 3D structure generation ... 

In this section, most of the currently available programs for automatic 3D structure generation will be discussed as far as they have been described in the literature. In addition, some early precursors of these methods are briefly presented due to their pioneering role in this field. 

The missing link between the constitution of a molecule and its 3D structure in computahonal chemistry is a technique capable of automatically generating 3D models starhng from the connectivity information of a given molecule. Due to its basic role, 3D structure generation is one of the fundamental problems in computahonal chemistry. As a consequence, in recent years a number of automahc 3D model builders and conformer generators have become available. For two comprehensive reviews, see Refs. [3, 4]. 

In this article, a classification of the specific concepts of different approaches to 3D-structure generation is undertaken and the domain covered in this article is defined. Under the wording automatic 3D-model builder we count programs capable of automatically predicting a 3D-molecular stnicture directly from the 2D-connectivity information and without user interaction. Most of the methods presented here are designed... 

High quality is one of the criteria defined in the requirements section above. Since the program should run automatically in batch mode, we mean by quality control an internal check of the 3D structures produced by the structure generator itself. In general, the abilities of a fast, automatic structure builder to assess the quaUty of its models are rather limited since, for example, an exhaustive conformation analysis and energy optimization is impossible in most cases. However, there are a Umited number of simple quaUty checks to avoid trivially distorted structures ... 

The 3D structure of the substrate to be analyzed (the starting conformation) has an impact on the outcome of the method. Satisfactory results were obtained using the in-house conformer generation, which is biased by the MIFs and the flexible shape of the active site of the enzymes. The latter procedure is automatically performed when a molecule or a set of molecules are provided in SMILE, 2D SDF or 3D co-ordinates. 

A special attention is given to stereochemistry, as some compounds are published without proper chirality representation even though the information is available, for example, for natural compounds and their derivatives. Furthermore, as illustrated in Fig. 13.2-8, compounds published in medicinal chemistry literature are often depicted in a human-readable format that is, structures are drawn in a format that chemists can interpret to reconstruct proper chirality. However, this format is not machine-readable , that is, cheminformatics software for 3D structural conversion, or for automatically generating IUPAC (International Union of Pure and Applied Chemistry) nomenclature, cannot perceive the stereo centers correctly... 

---