A branch-and-bound method for optimal atom-type assignment in de novo ligand design

Has Artificial Intelligence Impacted Drug Discovery?

Artificial intelligence (AI) tools find increasing application in drug discovery supporting every stage of the Design-Make-Test-Analyse (DMTA) cycle. The main focus of this chapter is the application in molecular generation with the aid of deep neural networks (DNN). We present a historical overview of the main advances in the field. We analyze the concepts of distribution and goal-directed learning and then highlight some of the recent applications of generative models in drug design with a focus into research work from the biopharmaceutical industry. We present in some more detail REINVENT which is an open-source software developed within our group in AstraZeneca and the main platform for AI molecular design support for a number of medicinal chemistry projects in the company and we also demonstrate some of our work in library design. Finally, we present some of the main challenges in the application of AI in Drug Discovery and different approaches to respond to these challenges which define areas for current and future work.

Fragment-based design of kinase inhibitors: a practical guide

Fragment-based drug design has become an important strategy for drug design and development over the last decade. It has been used with particular success in the development of kinase inhibitors, which are one of the most widely explored classes of drug targets today. The application of fragment-based methods to discovering and optimizing kinase inhibitors can be a complicated and daunting task; however, a general process has emerged that has been highly fruitful. Here a practical outline of the fragment process used in kinase inhibitor design and development is laid out with specific examples. A guide to the overall process from initial discovery through fragment screening, including the difficulties in detection, to the computational methods available for use in optimization of the discovered fragments is reported.

Integrating structure-based and ligand-based approaches for computational drug design

Methods utilized in computer-aided drug design can be classified into two major categories: structure based and ligand based, using information on the structure of the protein or on the biological and physicochemical properties of bound ligands, respectively. In recent years there has been a trend towards integrating these two methods in order to enhance the reliability and efficiency of computer-aided drug-design approaches by combining information from both the ligand and the protein. This trend resulted in a variety of methods that include: pseudoreceptor methods, pharmacophore methods, fingerprint methods and approaches integrating docking with similarity-based methods. In this article, we will describe the concepts behind each method and selected applications.

Fragment-based drug design

Fragment-based drug design (FBDD), which is comprised of both fragment screening and the use of fragment hits to design leads, began more than 15 years ago and has been steadily gaining in popularity and utility. Its origin lies on the fact that the coverage of chemical space and the binding efficiency of hits are directly related to the size of the compounds screened. Nevertheless, FBDD still faces challenges, among them developing fragment screening libraries that ensure optimal coverage of chemical space, physical properties and chemical tractability. Fragment screening also requires sensitive assays, often biophysical in nature, to detect weak binders. In this chapter we will introduce the technologies used to address these challenges and outline the experimental advantages that make FBDD one of the most popular new hit-to-lead process.

De novo design: balancing novelty and confined chemical space

IMPORTANCE OF THE FIELD

De novo drug design serves as a tool for the discovery of new ligands for macromolecular targets as well as optimization of known ligands. Recently developed tools aim to address the multi-objective nature of drug design in an unprecedented manner.

AREAS COVERED IN THIS REVIEW

This article discusses recent advances in de novo drug design programs and accessory programs used to evaluate compounds post-generation.

WHAT THE READER WILL GAIN

The reader is introduced to the challenges inherent in de novo drug design and will become familiar with current trends in de novo design. Furthermore, the reader will be better prepared to assess the value of a tool, and be equipped to design more elegant tools in the future.

TAKE HOME MESSAGE

De novo drug design can assist in the efficient discovery of new compounds with a high affinity for a given target. The inclusion of existing chemoinformatic methods with current structure-based de novo design tools provides a means of enhancing the therapeutic value of these generated compounds.

FlexNovo: structure-based searching in large fragment spaces

We present a new molecular design program, FlexNovo, for structure-based searching within large fragment spaces following a sequential growth strategy. The fragment spaces consist of several thousands of chemical fragments and a corresponding set of rules that specify how the fragments can be connected. FlexNovo is based on the FlexX molecular docking software and makes use of its incremental construction algorithm and the underlying chemical models. Interaction energies are calculated by using standard scoring functions. Several placement geometry, physicochemical property (drug-likeness), and diversity filter criteria are directly integrated into the "build-up" process. FlexNovo has been used to design potential inhibitors for four targets of pharmaceutical interest (dihydrofolate reductase, cyclin-dependant kinase 2, cyclooxygenase-2, and the estrogen receptor). We have carried out calculations using different diversity parameters for each of these targets and generated solution sets containing up to 50 molecules. The compounds obtained show that FlexNovo is able to generate a diverse set of reasonable molecules with drug-like properties. The results, including an automated similarity analysis with the Feature Tree program, indicate that FlexNovo often reproduces structural motifs as well as the corresponding binding modes seen in known active structures.

Small-Molecule Inhibitors of the MDM2-p53 Protein−Protein Interaction Based on an Isoindolinone Scaffold

From a set of weakly potent lead compounds, using in silico screening and small library synthesis, a series of 2-alkyl-3-aryl-3-alkoxyisoindolinones has been identified as inhibitors of the MDM2-p53 interaction. Two of the most potent compounds, 2-benzyl-3-(4-chlorophenyl)-3-(3-hydroxypropoxy)-2,3-dihydroisoindol-1-one (76; IC(50) = 15.9 +/- 0.8 microM) and 3-(4-chlorophenyl)-3-(4-hydroxy-3,5-dimethoxybenzyloxy)-2-propyl-2,3-dihydroisoindol-1-one (79; IC(50) = 5.3 +/- 0.9 microM), induced p53-dependent gene transcription, in a dose-dependent manner, in the MDM2 amplified, SJSA human sarcoma cell line.

Exhaustive de novo design of low-molecular-weight fragments against the ATP-binding site of DNA-gyrase

We present a de novo design approach to generating small fragments in the DNA-gyrase ATP-binding site using the computational drug design platform SkelGen. We have generated an exhaustive number of structural possibilities, which were subsequently filtered for site complementarity and synthetic tractability. A number of known active fragments are found, but most of the species created are potentially novel and could be valuable for further elaboration and development into lead-like structures.

The influence of variations of ligand protonation and tautomerism on protein-ligand recognition and binding energy landscape

We investigate the influence of variations of ligand protonation and tautomeric states on the protein-ligand binding energy landscape by applying the concept of structural consensus. In docking simulations, allowing full flexibility of the ligand, we explore whether the native binding mode could be successfully recovered using a non-native ligand protonation state. Here, we consider three proteins, dihydrofolate reductase, transketolase, and alpha-trichosanthin, complexed with ligands having multiple tautomeric forms. We find that for the majority of protonation and tautomeric states the native binding mode can be recovered without a great loss of accuracy.

The effect of a tightly bound water molecule on scaffold diversity in the computer-aided de novo ligand design of CDK2 inhibitors

We have determined the effects that tightly bound water molecules have on the de novo design of cyclin-dependent kinase-2 (CDK2) ligands. In particular, we have analyzed the impact of a specific structural water molecule on the chemical diversity and binding mode of ligands generated through a de novo structure-based ligand generation method in the binding site of CDK2. The tightly bound water molecule modifies the size and shape of the binding site and we have found that it also imposed constraints on the observed binding modes of the generated ligands. This in turn had the indirect effect of reducing the chemical diversity of the underlying molecular scaffolds that were able to bind to the enzyme satisfactorily. [Figure: see text].

Generation and Selection of Novel Estrogen Receptor Ligands Using the De Novo Structure-Based Design Tool, SkelGen

A de novo design approach to generating novel estrogen receptor (ER) ligands is described. The SkelGen program was used to generate ligands in the active sites of seven crystal structures of ERalpha. Seventeen high-scoring, diverse structures were selected from the SkelGen output and synthesized without introducing any modifications to the structures. Five ligands, four of which are novel, showed < or = 25 microM affinity, with the best compound displaying an IC50 of 340 nM. SkelGen can, therefore, be a powerful tool for designing active molecules.

Computer-based de novo design of drug-like molecules

Ever since the first automated de novo design techniques were conceived only 15 years ago, the computer-based design of hit and lead structure candidates has emerged as a complementary approach to high-throughput screening. Although many challenges remain, de novo design supports drug discovery projects by generating novel pharmaceutically active agents with desired properties in a cost- and time-efficient manner. In this review, we outline the various design concepts and highlight current developments in computer-based de novo design.

Validation and use of the MM-PBSA approach for drug discovery

The MM-PBSA approach has become a popular method for calculating binding affinities of biomolecular complexes. Published application examples focus on small test sets and few proteins and, hence, are of limited relevance in assessing the general validity of this method. To further characterize MM-PBSA, we report on a more extensive study involving a large number of ligands and eight different proteins. Our results show that applying the MM-PBSA energy function to a single, relaxed complex structure is an adequate and sometimes more accurate approach than the standard free energy averaging over molecular dynamics snapshots. The use of MM-PBSA on a single structure is shown to be valuable (a) as a postdocking filter in further enriching virtual screening results, (b) as a helpful tool to prioritize de novo design solutions, and (c) for distinguishing between good and weak binders (DeltapIC(50) > or = 2-3), but rarely to reproduce smaller free energy differences.

Including Tightly-Bound Water Molecules in de Novo Drug Design. Exemplification through the in Silico Generation of Poly(ADP-ribose)polymerase Ligands

Different strategies for the in silico generation of ligand molecules in the binding site of poly(ADP-ribose)polymerase (PARP) were studied in order to observe the effect of the targeting and displacement of tightly bound water molecules. Several molecular scaffolds were identified as having better interactions in the binding site when targeting one or two tightly bound water molecules in the NAD binding site. Energy calculations were conducted in order to assess the ligand-protein and ligand-water-protein interactions of different functional groups of the generated ligands. These calculations were used to evaluate the energetic consequences of the presence of tightly bound water molecules and to identify those that contribute favorably to the binding of ligands.

Expanded Interaction Fingerprint Method for Analyzing Ligand Binding Modes in Docking and Structure-Based Drug Design

An expanded interaction fingerprint method has been developed for analyzing the binding modes of ligands in docking and structure-based design methods. Taking the basic premise of representing a ligand in terms of a binary string that denotes its interactions with a target protein, we have expanded the method to include additional interaction-specific information. By considering the hydrogen-bonding strength and/or accessibility of the hydrogen bonding groups within a binding site as well as their geometric arrangement we aim to provide a better representation of a ligand-protein interaction. These expanded methods have been applied to the postprocessing of binding poses generated in a docking study for 220 different proteins and to the analysis of ligands generated by an automated ligand-generation algorithm for the anthrax oedema factor. In the docking study, the application of the interaction fingerprint method as a postprocessing tool resulted in an increased success rate in identifying the crystallographic binding mode. In the analysis of the ligands generated for the anthrax oedema factor, the incorporation of additional interaction-specific information resulted in a more intuitive and comprehensive analysis of automated ligand-generation output.

Receptor Flexibility in the in Silico Screening of Reagents in the S1‘ Pocket of Human Collagenase

A major difficulty in structure-based molecular design is the prediction of the structure of the protein-ligand complex because of the enormous number of degrees of freedom. Commonly, the target protein is kept rigid in a single low-energy conformation. However, this does not reflect the dynamic nature of protein structures. In this work, we investigate the influence of receptor flexibility in virtual screening of reagents on a common scaffold in the S1' pocket of human collagenase (matrix metalloproteinase-1). We compare screening using a single-crystal structure and multiple NMR structures, both apo and holo forms. We also investigate two computational methods of addressing receptor flexibility that can be used when NMR data are not available. The results from virtual screening using the experimental structures are compared to those obtained using the two computational methods. From the results, we draw conclusions about the impact of target flexibility on the identification of active and diverse reagents in a virtual screening protocol.

A validation study on the practical use of automated de novo design

The de novo design program Skelgen has been used to design inhibitor structures for four targets of pharmaceutical interest. The designed structures are compared to modeled binding modes of known inhibitors (i) visually and (ii) by means of a novel similarity measure considering the size and spatial proximity of the maximum common substructure of two small molecules. It is shown that the Skelgen algorithm generates representatives of many inhibitor classes within a very short time and that the new similarity measure is useful for comparing and clustering designed structures. The results demonstrate the necessity of properly defining search constraints in practical applications of de novo design.

PARM: a practical utility for drug design

To accommodate situations in which the 3D structure of the target receptor is not available, we have developed the Pseudo Atomic Receptor Model (PARM) software package. In this article we describe PARM and illustrate its use with three examples: elemenes (potential anticancer drugs), angiotensin converting enzyme inhibitors, and human HIV-1 inhibitors TTD (1,1,3-trioxo-2H, 4H-thieno[3,4-e][1,2,4] thiadiazine derivatives). The results show that PARM can build models with favorable cross-validation statistics (Rcv2 values 0.7-0.9) and give helpful SAR insight. PARM has certain advantages: (a) it can be used for many systems, regardless of whether the 3D structure of the receptor is known; (b) PARM models were demonstrated to be highly statistically reliable; and (c) PARM analyses are robust and reproducible.

Synergy between combinatorial chemistry and de novo design

Traditional de novo design algorithms are able to generate many thousands of ligand structures that meet the constraints of a protein structure, but these structures are often not synthetically tractable. In this article, we describe how concepts from structure-based de novo design can be used to explore the search space in library design. A key feature of the approach is the requirement that specific templates are included within the designed structures. Each template corresponds to the "central core" of a combinatorial library. The template is positioned within an acyclic chain whose length and bond orders are systematically varied, and the conformational space of each structure that results (core plus chain) is explored to determine whether it is able to link together two or more strongly interacting functional groups or pharmacophores located within a protein binding site. This fragment connection algorithm provides "generic" 3D molecules in the sense that the linking part (minus the template) is built from an all-carbon chain whose synthesis may not be easily achieved. Thus, in the second phase, 2D queries are derived from the molecular skeletons and used to identify possible reagents from a database. Each potential reagent is checked to ensure that it is compatible with the conformation of its parent 3D conformation and the constraints of the binding site. Combinations of these reagents according to the combinatorial library reaction scheme give product molecules that contain the desired core template and the key functional/pharmacophoric groups, and would be able to adopt a conformation compatible with the original molecular skeleton without any unfavorable intermolecular or intramolecular interactions. We discuss how this strategy compares with and relates to alternative approaches to both structure-based library design and de novo design.

Branch-and-terminate: a combinatorial optimization algorithm for protein design

BACKGROUND

Several deterministic and stochastic combinatorial optimization algorithms have been applied to computational protein design and homology modeling. As structural targets increase in size, however, it has become necessary to find more powerful methods to address the increased combinatorial complexity.

RESULTS

We present a new deterministic combinatorial search algorithm called 'Branch-and-Terminate' (B&T), which is derived from the Branch-and-Bound search method. The B&T approach is based on the construction of an efficient but very restrictive bounding expression, which is used for the search of a combinatorial tree representing the protein system. The bounding expression is used both to determine the optimal organization of the tree and to perform a highly effective pruning procedure named 'termination'. For some calculations, the B&T method rivals the current deterministic standard, dead-end elimination (DEE), sometimes finding the solution up to 21 times faster. A more significant feature of the B&T algorithm is that it can provide an efficient way to complete the optimization of problems that have been partially reduced by a DEE algorithm.

CONCLUSIONS

The B&T algorithm is an effective optimization algorithm when used alone. Moreover, it can increase the problem size limit of amino acid sidechain placement calculations, such as protein design, by completing DEE optimizations that reach a point at which the DEE criteria become inefficient. Together the two algorithms make it possible to find solutions to problems that are intractable by either algorithm alone.