We present LoFT, a new approach for focused combinatorial library design. In contrast to existing methods, chemical fragment spaces, which mainly consist of a collection of fragments and connection rules, are used as the underlying search space. Selecting one or several core fragments with the same link pattern, a focused library can be designed.

LoFT combines classical physicochemical design criteria with the feature tree descriptor for similarity/dissimilarity measurement. By applying the comparison directly on fragment level, we are able to design focused libraries efficiently without explicitly combining the fragments. Several stochastic algorithms are provided for traversing the search space, employing a weighted multi-objective scoring function, filtering rules and diversity mechanisms. Besides e.g. simulated annealing and threshold acceptance, a cherry picking, selecting the n best products from the search space, is available.

For validation, LoFT was applied to several drug design scenarios. Starting with known drug molecules, we generated focused libraries within desired property ranges.

FlexNovo is a molecular design program for structure-based de novo searching in large fragment spaces following a sequential growth strategy. Having the active site as structural information, it uses fragment spaces as input that consist of several thousands of chemical fragments and a corresponding set of rules, which primarily specifies how the fragments can be connected with each other. Synthesizability can be ensured by several placement geometry, drug-likeness and diversity filter criteria that are directly integrated in the build-up process.

FlexNovo can be used for fragment expansion, e.g., starting from an Xray structure that has been produced in a fragment screen. Or it can be used entirely in a de novo fashion where the algorithm places fragments arbitrarily in the pocket and then grows the compound from the most promising ones.

We demonstrate the performance of FlexNovo on a few relevant medicinal chemistry projects.

Despite the good quality in uploaded protein structures, there are still uncertainties when it comes to defining an active site. The correct physico-chemical surrounding of a ligand can be crucial though in molecular docking and screening. E.g. defining metal atoms as pharmacophores, assigning an alternative amino acid and protonation state and the correct insertion and orientation of (displaceable) water molecules all play a vital role in the preparation of an active site for docking. Also, it might be necessary to scale the contribution of certain interactions to the overall score.
We show recent advances to prepare an active site for docking and screening and show a few proof-of-concepts.

Today, it is agreed that one major challenge in FBLD has emerged to be the synthetic accessibility of the products. However, another less publicly discussed, yet silently accepted boundary condition is pharma's need to protect intellectual property (IP). For several years now we have pursued tile-and-merge approaches. In this context we are left with two sub-problems: a) the search space generation (i.e., what are our fragments, their origin, and their connection rules) and b) how to search that space efficiently. In the framework of the above-mentioned, overall boundary conditions (synthetic accessibility and IP protection) we will discuss approaches to a) and b) which - prospectively - have led to remarkable success in big pharma. A recent one exploits the synergy between a ligand based search technique (FTrees) and a shape based filtering approach (ROCS). Along the way, technical obstacles plus how most of those could be circumvented will be shown.

FlexNovo is a molecular design program for structure-based de novo searching in large fragment spaces following a sequential growth strategy. Synthetic access can be ensured by several placement geometry, drug-likeness and diversity filter criteria that are directly integrated in the build-up process. FlexNovo can be used for fragment expansion, e.g., starting from an Xray structure that has been produced in a fragment screen. Or it can be used entirely in a de novo fashion where the algorithm places fragments arbitrarily in the pocket and then grows the compound from the most promising ones. We demonstrate the performance of FlexNovo on a few relevant medicinal chemistry projects.

Structure- and ligand-based virtual screening are very much in the heart of lead discovery efforts at early stages. Since three-dimensional molecular properties like the arrangement of pharmacophore features and the shape determine a molecule's ability to bind to a protein target of interest, the question arise, how molecules should be represented in screening databases.

In this talk, we introduce a novel molecular descriptor representing potential pharmacophoric points and shape such that they are suited for indexed database access. The descriptor can be combined with recently developed compressed bitmap indices to allow a highly efficient access to compounds based on structural features. The realization within structure-based and ligand-based virtual screening tools together with example applications will be presented.

Although computer-aided molecular design and virtual screening software tools improve continuously, manual investigation of the resulting complexes a control task in modelling. In contrast to 3D visualization, information contained in 2D plots can be identified by a short glance and are therefore more appropriate for scanning through large datasets.

We present a new version of PoseView^1,2 a computational method for the automatic generation of two-dimensional protein-ligand complex diagrams. The layout is computed considering hydrophilic, hydrophobic and metal contacts between ligand and receptor. While the ligand and protein residues forming hydrophilic interactions to the ligand are drawn according to chemical structure diagram conventions, the hydrophobic contacts are visualized by means of splines around the ligand and the appropriate residue labels. PoseView is based on a combinatorial layout optimization strategy which solves parts of the problem nonheuristically. The computation is performed in a sequential manner: An initial ligand structure diagram is created and subsequently modified in order to find a non-intersecting arrangement of interaction lines. In the following the initial placement of each hydrophilic interacting amino acid is computed. During the placement collisions are resolved by a branch & bound algorithm selecting an optimal relative arrangement of all amino acids and the ligand. Finally, the remaining components of the complex diagram are placed based on an underlying arrangement grid.

For validation, PoseView was applied to the protein-ligand complexes contained in the Brookhaven PDB database. Advantages and limitations of the approach will be discussed by means of representative test cases.

For examples see www.zbh.uni-hamburg.de/poseview.

Literature:

1. Stierand, K., Maaß, P., Rarey, M. (2006) Molecular Complexes at a Glance: Automated Generation of two-dimensional Complex Diagrams. Bioinformatics, 22, 1710-1716.
2. Stierand, K., Rarey, M. (2007). From Modeling to Medicinal Chemistry: Automatic Generation of Two-Dimensional Complex Diagrams. ChemMedChem 2, 6, 853-860.

Computational methods for lead identification and library design are frequently applied in cheminformatics. In case that structures should be designed de novo, chemical fragment spaces which result either from retrosynthetic breakdown of (drug-like) compounds or from analysis and combinatorial synthesis protocols substantially improve the drug-likeness and synthetic accessibility of the resulting compounds. Several cheminformatics tasks which are traditionally defined over individual molecules can also be formulated for chemical fragment spaces. Since the enumeration of all individual molecules of such a space is prohibitive, the challenge is to find computational methods solving modeling tasks without this step.

In this talk, we overview several methods developed for chemical fragment spaces. Among others, computational approaches to fragment space modeling and searching with various search criteria (property-, structure-, and similarity-based) are presented. Recently, we developed a novel method for deriving focused libraries from fragment spaces. Besides multiple physico-chemical property ranges, the similarity and dissimilarity to query molecules as well as the internal diversity can be considered. Some practical examples will demonstrate the capabilities as well as the limitations of these approaches.

Today, the primary domain of VS applications are screening collections based on in-house repositories and vendor collections. However, any such library is tiny compared to the number of synthetically accessible compounds from validated chemistries available in any pharma company. It would be of great interest to perform searches against such "virtual chemistry spaces". However, the number of possible compounds easily exceeds - by many orders of magnitude - the number of compounds that can be stored and searched using conventional methods.

We overcame these limitations by converting large numbers of existing (wet) combinatorial libraries into the now publicly available KnowledgeSpace covering 23 billion synthetically accessible compounds. FTrees - a fuzzy similarity calculator - is capable of searching such spaces within a few minutes. The result is a set of compounds similar to a query structure, plus an annotation through which of the synthetic routes these compounds can be made. FTrees is known for its scaffold hopping capabilities, thus chemically diverse results can be expected. Such output provides library design ideas for hit follow-up from high-throughput screening or lead hopping into novel series.

We present the design of the KnowledgeSpace and similar inhouse chemistry spaces as well as validation results and a number of successful applications including prospective results.

Fragments experience a buzz these days: Upon detection of fragment binders in protein active sites, the general strategy is to merge, grow, or link them to enhance binding of a resulting 'composite' ligand. Computational chemistry ideally supports this workflow by sensible proposals for synthesis and modifications. However, on the computational side complications are the lack of time and the quality of checks for synthetic accessibility of the proposals.

Our tool ReCore was extended to identify linker motifs from excessively large fragment libraries which do not only connect fragment binders in their experimentally observed position, but also comply with the binding motifs using pharmacophores and other features.

ReCore is fast enough to provide instant feedback to the user - thereby enabling an interactive query refinement. The algorithm moreover favors synthetically feasible solutions by the setup of its search libraries and upon forming the resulting composite ligands. Validations across different targets are reported.

An often encountered scenario in FBLD is experimental evidence of one or multiple fragment binders in a protein binding site. Typically, depending on quality and amount of information, subsequent steps can be divided into three classes: a) growing from these 'seeds', b) linking of two or more fragment binders, or c) merging multiple overlapping binders into a single potent lead.

To accomplish these tasks with ultimate efficiency, the software tool ReCore has been developed. Based on a vast 3D fragment library, ReCore finds fragments which provide an optimal fit with the 'dangling bonds' and comply with optional filters and pharmacophore features. Based on a novel indexing technology, ReCore, in contrast to other tools targeting a similar challenge, provides its results within seconds, thus allowing interactive usage.

Synthetic access of results (which was a major weakness in the early days of de novo design) is taken care of in ReCore at three levels: during fragment creation, within query definition, and when creating the results. We will elucidate the basic principles and give examples which map onto experimental data and evolve into novel lead ideas.

Virtual screening is widely established as part of the drug discovery process. So far, the primary domain of application are screening collections based on in-house repositories and vendor collections, while the pharma companies have access to large numbers of validated chemistries. It would be of great interest to perform similarity searches against all compounds that are synthetically accessible by any such combinatorial library protocol. However, the number of possible compounds easily exceeds – by many orders of magnitude – the number of compounds that can be stored and searched by conventional methods.

We overcame these limitations by converting large numbers of combinatorial libraries into a "virtual chemistry space". FTrees – a similarity calculator – is capable of searching such spaces without ever enumerating all possible molecular structures. It produces a set of compounds similar to a query structure, which are synthetically accessible via one or more of the existing synthetic routes. FTrees is known for its scaffold-hopping capabilities, thus chemically diverse results can be expected. Such output provides library design ideas for hit follow-up from high-throughput screening or lead hopping into novel series. A number of successful applications will be presented.

Fragment-Based Ligand Design (FBLD) has shown that one major challenge occurs after the identifcation of a small fragment binder in a binding pocket: How to grow from one intial, binding fragment into other parts of the pocket, so that a favorably binding, lead-like molecule can be formed?

Until now, modellers have had only a small choice of tools supporting this scenario. FlexSISNovo is the first structure-based tool which is specially designed to grow from an experimentally observed ligand structure and uses so-called fragment spaces as an input to ensure synthetic access of the resulting molecules at the same time. The input fragment spaces can be created either from sensibly shredding compounds and/or by using combinatorial chemistry protocols.

Before, during, and after docking, further knowledge, e.g., about necessary interactions, logP constraints, mass tolerances and other filter criteria can be taken into account.

The talk will document on the technology we implemented and discuss application examples using a novel, public domain fragment space created from real combinatorial chemistry protocols.

Virtual high throughput screening of in-house compound collections and vendor catalogs is a validated approach in the quest for novel molecular entities. However, these libraries are small compared to the overall synthesizable number of compounds from validated "wet" chemical reactions in pharma companies or the public domain.

In order to overcome this limitation, we designed a large virtual combinatorial chemistry space from publicly available combinatorial libraries that gives access to billions of synthetically accessible compounds. Together with FTrees, a fuzzy similarity calculator, the researcher has a means of searching this KnowledgeSpace for analogues to one or several query molecules within a few minutes. The resulting compounds not only exhibit similar properties to the query molecule(s), but also feature an annotation through which of the synthetic routes these molecules can be made. Results can be expected to be diverse, based on FTrees scaffold hopping capabilities, and provide ideas for hit follow-up into novel compound classes.

In this contribution we present the design and properties of the KnowledgeSpace and other in-house chemistry spaces that build on the same strategy as well as validation of results and a number of successful applications including prospective results.