FlexNovo: Structure-Based Searching in Large Fragment Spaces

Jörg Degen, Matthias Rarey

The full details of the FlexNovo software package and the initial validation experiments can be found in the original publication: ChemMedChem 2006, 1, 854-868. [DOI]

Introduction

At the beginning of a drug discovery project, the coverage of a reasonable amount of the chemical space is a crucial point for the identification of possible lead structures for the target of interest. Until now, mostly increasingly large experimental and in silico screening collections have been used for addressing this problem. However, the amount of chemical space that these methods can cover is quite small compared to the amount of compounds available in principle. One possible alternative is to use fragment spaces instead of compound libraries [1-3]. Since such spaces are many orders of magnitude too large to be enumerated, new modeling tools are needed that exploit the combinatorial structure of these spaces. FlexNovo is a new structure-based molecular design software developed to address this need [4].

Fragment Spaces

A fragment space consists of chemical fragments, being its 'atomic' elements. The figure below shows an example collection of fragment prototypes contained in such a fragment space. In general, a fragment has one or several open valences, represented by so called link (dummy) atoms. Each of these link atoms has a certain type representing different chemical environments (L1-L12 in the image below). The fragments span the fragment space through the definition of the compatibility of link types (indicated by green lines in the image below). Fragments can be connected via the formation of a bond between the atoms adjacent to the link atoms. The link atoms themselves will be removed upon fragment connection. Finally, for each link type, a so-called terminal group has to be specified, which is meant to substitute the corresponding link atom in case it has not been involved in fragment connection.

FlexNovo algorihm

FlexNovo is based on the FlexX [5] molecular docking software and makes use of its incremental construction algorithm and the underlying chemical models. Interaction energies are calculated using standard scoring functions. Several placement geometry, physicochemical property (drug-likeness), and diversity filter criteria are directly integrated into the build-up process. FlexNovo also uses the Pharm-extension [6] of FlexX that allows for the consideration of pharmacophore-type constraints.

The image below contains a brief summary of the main computational steps and shows the tight integration of the filter criteria in the build-up calculation. A more detailed description of the algorithm can be found in [4].

The result of a FlexNovo calculation is a collection of high-scoring compounds, which meet certain physicochemical property and diversity criteria. For each molecule contained in this collection, a number of different orientations (and conformations) in the active site are generated.

FlexNovo results

FlexNovo has been used to design potential inhibitors for four targets of pharmaceutical interest: dihydrofolate reductase (DHFR), cyclin-dependant kinase 2 (CDK2), cyclooxygenase-2 (COX-2) and estrogen receptor (ER). We have carried out calculations for each of these targets and generated solution sets containing up to 50 molecules. The following figure shows one representative example for each application scenario resulting from a typical calculation with standard parameters. The individual images contain the corresponding solution drawn according to the best-predicted binding orientation together with the Connolly surface of the active site of the protein. The top half of the cavity is removed. Surface patches resulting from heteroatoms are shown in the corresponding colors.

FlexNovo solution for DHFR
FlexNovo solution for CDK2
FlexNovo solution for ER
FlexNovo solution for COX-2

We analyzed the solution sets derived from the calculations described above regarding their distributions for a set of relevant physicochemical properties. We compared these to the corresponding distributions derived from a set of known inhibitors. The following figure shows two selected distributions for molecular weight (left image) and calculated log P values (right image). Histograms are normalized and represented as smoothed line graphs. Each individual image contains the superimposed distributions for the FlexNovo solutions (solid lines) and known actives (dashed lines) for the four different targets: DHFR (dark blue), CDK2 (red), COX-2 (yellow) and ER (green).

Summary

Searching a fragment space with about 17000 fragments can be done on a standard workstation computer, takes less than one day of computation time and needs 750 MB of memory. The fragment space can be defined such that easily accessible fragments (building blocks) and well defined synthesis steps (connection rules) are considered. The compounds obtained show that FlexNovo is able to generate diverse sets of reasonable molecules being highly complementary to different target proteins and fulfilling drug-like properties to a certain extent. By comparing these to known inhibitors, similarities with respect to their structures and binding modes are frequently observed. Because of this, we think that FlexNovo is a valuable tool for a structure-based design objective.

FlexNovo is meant to be an integrated and user-customizable idea-generating system. It may help in suggesting new structural motifs or even in designing a first library for a given target protein and help in understanding interactions and complementarity issues in the context of protein-ligand binding. The algorithmic and conceptual improvements we plan might further increase the potency of the current approach.

Acknowledgements

This work is associated with the NovoBench project funded under grant 313324A by the bmb+f in the framework of the BioChance+ program. Further cooperation partners are: 4SC AG (Martinsried, Germany), ALTANA Pharma AG (Konstanz, Germany), BioSolveIT GmbH (Sankt Augustin, Germany), Computer Chemistry Center (University of Erlangen-Nuremberg, Germany), Lilly Research Laboratories (Hamburg, Germany) and Molecular Networks GmbH (Erlangen, Germany).

Further Information and Software Availability

The FlexNovo system is under ongoing and further development in cooperation with BioSolveIT. Evaluation licenses can be made available upon request by BioSolveIT GmbH. For further information or questions concerning the FlexNovo technology please contact Jörg Degen or Matthias Rarey.

The FlexX program package is developed and distributed by BioSolveIT GmbH.

Literature

[1] Lewell, X. Q., Judd, D. B., Watson, S. P., Hann, M. M. (1998). RECAP - Retrosynthetic Combinatorial Analysis Procedure: A Powerful New Technique for Identifying Privileged Molecular Fragments with Useful Applications in Combinatorial Chemistry. Journal of Chemical Information and Computer Sciences, 38, 511-522. [DOI]

[2] Schneider, G., Lee, M.-L., Stahl, M., Schneider, P. (2000). De novo design of molecular architectures by evolutionary assembly of drug-derived building blocks. Journal of Computer-Aided Molecular Design, 14, 487-494. [DOI]

[3] Rarey, M., Stahl, M. (2001). Similarity searching in large combinatorial chemistry spaces. Journal of Computer-Aided Molecular Design, 15, 497-520. [DOI]

[4] Degen, J., Rarey, M. (2006). FlexNovo: Structure-Based Searching in Large Fragment Spaces. ChemMedChem, 1, 854-868. [DOI]

[5] Rarey, M., Kramer, B., Lengauer, T., Klebe, G. (1996). A Fast Flexible Docking Method using an Incremental Construction Algorithm. Journal of Molecular Biology 261(3), 470-489. [DOI]

[6] Hindle S. A., Rarey M., Bunning C., Lengauer T. (2002). Flexible Docking under Pharmacophore Type Constraints. Journal of Computer-Aided Molecular Design, 16, 129-149. [DOI]