A genetic algorithm for the automated generation of molecules within constraints

Summary A genetic algorithm has been designed which generates molecular structures within constraints. The constraints may be any useful function, for example an enzyme active site, a pharmacophore or molecular properties from pattern recognition or rule-induction analyses. The starting point may be random or may utilise known molecules. These are modified to grow into families of structures which, using the evolutionary operators of selection, crossover and mutation evolve to better fit the constraints. The basis of the algorithm is described together with some applications in lead generation, 3D database construction and drug design. Genetic algorithms of this type may have wider applications in chemistry, for example in the design and optimisation of new polymers, materials (e.g. superconducting materials) or synthetic enzymes.     

Automated docking using a Lamarckian genetic algorithm and an empirical binding free energy function

A novel and robust automated docking method that predicts the bound conformations of flexible ligands to macromolecular targets has been developed and tested, in combination with a new scoring function that estimates the free energy change upon binding. Interestingly, this method applies a Lamarckian model of genetics, in which environmental adaptations of an individual's phenotype are reverse transcribed into its genotype and become heritable traits (sic). We consider three search methods, Monte Carlo simulated annealing, a traditional genetic algorithm, and the Lamarckian genetic algorithm, and compare their performance in dockings of seven protein–ligand test systems having known three-dimensional structure. We show that both the traditional and Lamarckian genetic algorithms can handle ligands with more degrees of freedom than the simulated annealing method used in earlier versions of AUTO DOCK , and that the Lamarckian genetic algorithm is the most efficient, reliable, and successful of the three. The empirical free energy function was calibrated using a set of 30 structurally known protein–ligand complexes with experimentally determined binding constants. Linear regression analysis of the observed binding constants in terms of a wide variety of structure-derived molecular properties was performed. The final model had a residual standard error of 9.11 kJ mol⁻¹ (2.177 kcal mol⁻¹) and was chosen as the new energy function. The new search methods and empirical free energy function are available in AUTO DOCK , version 3.0. © 1998 John Wiley & Sons, Inc. J Comput Chem 19: 1639–1662, 1998     

CHARMM‐GUI ligand reader and modeler for CHARMM force field generation of small molecules

Reading ligand structures into any simulation program is often nontrivial and time consuming, especially when the force field parameters and/or structure files of the corresponding molecules are not available. To address this problem, we have developed Ligand Reader & Modeler in CHARMM‐GUI. Users can upload ligand structure information in various forms (using PDB ID, ligand ID, SMILES, MOL/MOL2/SDF file, or PDB/mmCIF file), and the uploaded structure is displayed on a sketchpad for verification and further modification. Based on the displayed structure, Ligand Reader & Modeler generates the ligand force field parameters and necessary structure files by searching for the ligand in the CHARMM force field library or using the CHARMM general force field (CGenFF). In addition, users can define chemical substitution sites and draw substituents in each site on the sketchpad to generate a set of combinatorial structure files and corresponding force field parameters for throughput or alchemical free energy simulations. Finally, the output from Ligand Reader & Modeler can be used in other CHARMM‐GUI modules to build a protein‐ligand simulation system for all supported simulation programs, such as CHARMM, NAMD, GROMACS, AMBER, GENESIS, LAMMPS, Desmond, OpenMM, and CHARMM/OpenMM. Ligand Reader & Modeler is available as a functional module of CHARMM‐GUI at http://www.charmm-gui.org/input/ligandrm . © 2017 Wiley Periodicals, Inc.     

A genetic algorithm for structure-based de novo design

Genetic algorithms have properties which make them attractive in de novo drug design. Like other de novo design programs, genetic algorithms require a method to reduce the enormous search space of possible compounds. Most often this is done using information from known ligands. We have developed the ADAPT program, a genetic algorithm which uses molecular interactions evaluated with docking calculations as a fitness function to reduce the search space. ADAPT does not require information about known ligands. The program takes an initial set of compounds and iteratively builds new compounds based on the fitness scores of the previous set of compounds. We describe the particulars of the ADAPT algorithm and its application to three well-studied target systems. We also show that the strategies of enhanced local sampling and re-introducing diversity to the compound population during the design cycle provide better results than conventional genetic algorithm protocols.     

Genetic algorithm for the design of molecules with desired properties

The design of molecules with desired properties is still a challenge because of the largely unpredictable end results. Computational methods can be used to assist and speed up this process. In particular, genetic algorithms have proved to be powerful tools with a wide range of applications, e.g. in the field of drug development. Here, we propose a new genetic algorithm that has been tailored to meet the demands of de novo drug design, i.e. efficient optimization based on small training sets that are analyzed in only a small number of design cycles. The efficiency of the design algorithm was demonstrated in the context of several different applications. First, RNA molecules were optimized with respect to folding energy. Second, a spinglass was optimized as a model system for the optimization of multiletter alphabet biopolymers such as peptides. Finally, the feasibility of the computer-assisted molecular design approach was demonstrated for the de novo construction of peptidic thrombin inhibitors using an iterative process of 4 design cycles of computer-guided optimization. Synthesis and experimental fitness determination of only 600 different compounds from a virtual library of more than 10¹⁷ molecules was necessary to achieve this goal.These authors contributed equally to the results presentedThese authors contributed equally to the results presentedThese authors contributed equally to the results presentedThese authors contributed equally to the results presented     

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

This paper investigates a computational procedure for the determination of the atom types on the vertices of a molecular skeleton to optimize interaction with the receptor site whilst maintaining a synthetically reasonable structure. The connectivity of the skeleton is analysed and appropriate atom types are compiled for each vertex. Receptor ionization and conformational states are generated by varying the positions of hydrogen atoms and electron lone pairs in the carboxyl, rotatable hydroxyl and amino groups. The structure is divided into small non-overlapping substructures. Atom types are assigned exhaustively onto each of the substructures using a depth-first search method; chemical rules are applied to reject unacceptable atom combinations early on. An empirical interaction score is calculated and the representatives of each partial structure are stored in ascending order according to their scores. The branch-and-bound procedure is then used to find the structures with the lowest scores. The method is illustrated using five protein–ligand complexes.     

CAVEAT: A program to facilitate the design of organic molecules

Summary A frequently encountered problem in the design of enzyme inhibitors and other biologically active molecules is the identification of molecular frameworks to serve as templates or linking units that can position functional groups in specific relative orientations. The program CAVEAT was designed to address this problem by searching 3D databases for such molecular fragments. Key innovations introduced in CAVEAT are a focus on relationships between bonds and the provision of automated methods to identify and classify structural frameworks. Performance has been a particular concern in formulating CAVEAT, since it is intended to be used in an interactive manner. The focus in this report is the design and implementation of the principal algorthms and the performance achieved.CAVEAT is available from the Office of Technology Licensing, University of California, Berkeley, CA, and from Molecular Simulations, Inc., Burlington, MA; further information is available from P.A. Bartlett.     

A toolchain for the automatic generation of computer codes for correlated wavefunction calculations

In this work, the automated generator environment for ORCA (ORCA‐AGE) is described. It is a powerful toolchain for the automatic implementation of wavefunction‐based quantum chemical methods. ORCA‐AGE consists of three main modules: (1) generation of “raw” equations from a second quantized Ansatz for the wavefunction, (2) factorization and optimization of equations, and (3) generation of actual computer code. We generate code for the ORCA package, making use of the powerful functionality for wavefunction‐based correlation calculations that is already present in the code. The equation generation makes use of the most elementary commutation relations and hence is extremely general. Consequently, code can be generated for single reference as well as multireference approaches and spin‐independent as well as spin‐dependent operators. The performance of the generated code is demonstrated through comparison with efficient hand‐optimized code for some well‐understood standard configuration interaction and coupled cluster methods. In general, the speed of the generated code is no more than 30% slower than the hand‐optimized code, thus allowing for routine application of canonical ab initio methods to molecules with about 500–1000 basis functions. Using the toolchain, complicated methods, especially those surpassing human ability for handling complexity, can be efficiently and reliably implemented in very short times. This enables the developer to shift the attention from debugging code to the physical content of the chosen wavefunction Ansatz. Automatic code generation also has the desirable property that any improvement in the toolchain immediately applies to all generated code. © 2017 Wiley Periodicals, Inc.     

CHARMM general force field: A force field for drug‐like molecules compatible with the CHARMM all‐atom additive biological force fields

The widely used CHARMM additive all‐atom force field includes parameters for proteins, nucleic acids, lipids, and carbohydrates. In the present article, an extension of the CHARMM force field to drug‐like molecules is presented. The resulting CHARMM General Force Field (CGenFF) covers a wide range of chemical groups present in biomolecules and drug‐like molecules, including a large number of heterocyclic scaffolds. The parametrization philosophy behind the force field focuses on quality at the expense of transferability, with the implementation concentrating on an extensible force field. Statistics related to the quality of the parametrization with a focus on experimental validation are presented. Additionally, the parametrization procedure, described fully in the present article in the context of the model systems, pyrrolidine, and 3‐phenoxymethylpyrrolidine will allow users to readily extend the force field to chemical groups that are not explicitly covered in the force field as well as add functional groups to and link together molecules already available in the force field. CGenFF thus makes it possible to perform “all‐CHARMM” simulations on drug‐target interactions thereby extending the utility of CHARMM force fields to medicinally relevant systems. © 2009 Wiley Periodicals, Inc. J Comput Chem, 2010     

De novo ligand design with explicit water molecules: an application to bacterial neuraminidase

Most computer-aided drug design methods ignore the presence of crystallographically-determined water molecules in the binding site of a target protein. In this paper, our de novo ligand design methods are applied to the X-ray crystal structure of bacterial neuraminidase in the presence of some selected water molecules. We have found that, for this particular protein, the complete removal of all bound water molecules leads to difficulties in generating any potential ligands if the unsatisfied hydrogen-bonding sitepoints left by removing these water molecules are to be satisfied by a ligand. As more of the crystallographically determined water molecules are allowed in the binding site, it becomes much easier to generate ligands in larger numbers and with wider chemical diversity. This example shows that, in some cases, bound water molecules can be more accessible for hydrogen bonding to an incoming ligand than the actual protein binding sitepoints associated with them. From the point of view of de novo ligand design, water molecules can thus act as versatile amphiprotic hydrogen-bonding sitepoints and reduce the conformational constraints of a particular binding site.     

Heuristic lipophilicity potential for computer-aided rational drug design: Optimizations of screening functions and parameters

In this research we test and compare three possible atom-basedscreening functions used in the heuristic molecular lipophilicity potential(HMLP). Screening function 1 is a power distance-dependent function, b , screening function 2is an exponential distance-dependent function, b_iexp( , and screening function 3 is aweighted distance-dependent function, For every screening function, the parameters ( ,d₀, and are optimized using 41 common organic molecules of 4 types of compounds:aliphatic alcohols, aliphatic carboxylic acids, aliphatic amines, andaliphatic alkanes. The results of calculations show that screening function3 cannot give chemically reasonable results, however, both the powerscreening function and the exponential screening function give chemicallysatisfactory results. There are two notable differences between screeningfunctions 1 and 2. First, the exponential screening function has largervalues in the short distance than the power screening function, thereforemore influence from the nearest neighbors is involved using screeningfunction 2 than screening function 1. Second, the power screening functionhas larger values in the long distance than the exponential screeningfunction, therefore screening function 1 is effected by atoms at longdistance more than screening function 2. For screening function 1, thesuitable range of parameter d₀ is 1.5 < d₀ < 3.0, and d₀ = 2.0 is recommended. HMLP developed in this researchprovides a potential tool for computer-aided three-dimensional drugdesign.     

RMC_POT: A computer code for reverse monte carlo modeling the structure of disordered systems containing molecules of arbitrary complexity

An approach has been devised and tested for preserving the molecular dynamics molecular geometry taking into account energetic considerations during Reverse Monte Carlo (RMC) modeling. Instead of the commonly used fixed neighbor constraints, where molecules are held together by constraining distance ranges available for the specified atom pairs, here molecules are kept together via bond, angle, and dihedral potential energies. The scaled total potential energy contributes to the measure of the goodness‐of‐fit, thus, the atoms can be prevented from drifting apart. In some of the calculations (Lennard‐Jones and Coulombic) nonbonding potentials were also applied. The algorithm was successfully tested for the X‐ray structure factor‐based structure study of liquid dimethyl trisulfide, for which material now significantly more sensible results have been obtained than during previous attempts via any earlier version of RMC modeling. It is envisaged that structural modeling of a large class of materials, primarily liquids and amorphous solids containing molecules of up to about 100 atoms, will make use of the new code in the near future. © 2012 Wiley Periodicals, Inc.     

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.     

Ab initio protein fold prediction using evolutionary algorithms: influence of design and control parameters on performance

True ab initio prediction of protein 3D structure requires only the protein primary structure, a physicochemical free energy model, and a search method for identifying the free energy global minimum. Various characteristics of evolutionary algorithms (EAs) mean they are in principle well suited to the latter. Studies to date have been less than encouraging, however. This is because of the limited consideration given to EA design and control parameter issues. A comprehensive study of these issues was, therefore, undertaken for ab initio protein fold prediction using a full atomistic protein model. The performance and optimal control parameter settings of twelve EA designs where first established using a 15-residue polyalanine molecule-design aspects varied include the encoding alphabet, crossover operator, and replacement strategy. It can be concluded that real encoding and multipoint crossover are superior, while both generational and steady-state replacement strategies have merits. The scaling between the optimal control parameter settings and polyalanine size was also identified for both generational and steady-state designs based on real encoding and multipoint crossover. Application of the steady-state design to met-enkephalin indicated that these scalings are potentially transferable to real proteins. Comparison of the performance of the steady state design for met-enkephalin with other ab initio methods indicates that EAs can be competitive provided the correct design and control parameter values are used.     

Multi‐objective optimization of heating channels for rapid heating cycle injection mold using Pareto‐based genetic algorithm

Rapid heating cycle molding (RHCM) technology is a novel polymer injection molding method developed in recent years. In this paper, the principle of RHCM technology was introduced and a RHCM mold for producing a large‐size LCD TV panel was presented. Aiming at achieving a uniform temperature distribution on the cavity surface as well as making sure of the heating efficiency, the factors that influence temperature distribution and heating efficiency were analyzed. The center coordinates of the heating channels were considered as the main design variables. Multi‐objective functions for optimizing the temperature distribution uniformity and heating efficiency were proposed. The layout of the heating channels for a 46‐inch LCD TV panel RHCM mold was optimized by using the finite element method and Pareto‐based genetic algorithm (GA). The temperature distribution uniformity on the stationary mold insert cavity surface was largely improved by using optimal design results and the heating efficiency was also guaranteed. Copyright © 2009 John Wiley & Sons, Ltd.     

Deep Learning and Structure-Based Virtual Screening for Drug Discovery against NEK7: A Novel Target for the Treatment of Cancer

NIMA-related kinase7 (NEK7) plays a multifunctional role in cell division and NLRP3 inflammasone activation. A typical expression or any mutation in the genetic makeup of NEK7 leads to the development of cancer malignancies and fatal inflammatory disease, i.e., breast cancer, non-small cell lung cancer, gout, rheumatoid arthritis, and liver cirrhosis. Therefore, NEK7 is a promising target for drug development against various cancer malignancies. The combination of drug repurposing and structure-based virtual screening of large libraries of compounds has dramatically improved the development of anticancer drugs. The current study focused on the virtual screening of 1200 benzene sulphonamide derivatives retrieved from the PubChem database by selecting and docking validation of the crystal structure of NEK7 protein (PDB ID: 2WQN). The compounds library was subjected to virtual screening using Auto Dock Vina. The binding energies of screened compounds were compared to standard Dabrafenib. In particular, compound 762 exhibited excellent binding energy of −42.67 kJ/mol, better than Dabrafenib (−33.89 kJ/mol). Selected drug candidates showed a reactive profile that was comparable to standard Dabrafenib. To characterize the stability of protein–ligand complexes, molecular dynamic simulations were performed, providing insight into the molecular interactions. The NEK7–Dabrafenib complex showed stability throughout the simulated trajectory. In addition, binding affinities, pIC50, and ADMET profiles of drug candidates were predicted using deep learning models. Deep learning models predicted the binding affinity of compound 762 best among all derivatives, which supports the findings of virtual screening. These findings suggest that top hits can serve as potential inhibitors of NEK7. Moreover, it is recommended to explore the inhibitory potential of identified hits compounds through in-vitro and in-vivo approaches.     

Properties of a genetic algorithm extended by a random self-learning operator and asymmetric mutations: A convergence study for a task of powder-pattern indexing

Genetic algorithms represent a powerful global-optimisation tool applicable in solving tasks of high complexity in science, technology, medicine, communication, etc. The usual genetic-algorithm calculation scheme is extended here by introduction of a quadratic self-learning operator, which performs a partial local search for randomly selected representatives of the population. This operator is aimed as a minor deterministic contribution to the (stochastic) genetic search. The population representing the trial solutions is split into two equal subpopulations allowed to exhibit different mutation rates (so called asymmetric mutation). The convergence is studied in detail exploiting a crystallographic-test example of indexing of powder diffraction data of orthorhombic lithium copper oxide, varying such parameters as mutation rates and the learning rate. It is shown through the averaged (over the subpopulation) fitness behaviour, how the genetic diversity in the population depends on the mutation rate of the given subpopulation. Conditions and algorithm parameter values favourable for convergence in the framework of proposed approach are discussed using the results for the mentioned example. Further data are studied with a somewhat modified algorithm using periodically varying mutation rates and a problem-specific operator. The chance of finding the global optimum and the convergence speed are observed to be strongly influenced by the effective mutation level and on the self-learning level. The optimal values of these two parameters are about 6 and 5%, respectively. The periodic changes of mutation rate are found to improve the explorative abilities of the algorithm. The results of the study confirm that the applied methodology leads to improvement of the classical genetic algorithm and, therefore, it is expected to be helpful in constructing of algorithms permitting to solve similar tasks of higher complexity.