Peptide design by artificial neural networks and computer-based evolutionary search

Determinants of Chromatin Organization in Aging and Cancer-Emerging Opportunities for Epigenetic Therapies and AI Technology

This review article critically examines the pivotal role of chromatin organization in gene regulation, cellular differentiation, disease progression and aging. It explores the dynamic between the euchromatin and heterochromatin, coded by a complex array of histone modifications that orchestrate essential cellular processes. We discuss the pathological impacts of chromatin state misregulation, particularly in cancer and accelerated aging conditions such as progeroid syndromes, and highlight the innovative role of epigenetic therapies and artificial intelligence (AI) in comprehending and harnessing the histone code toward personalized medicine. In the context of aging, this review explores the use of AI and advanced machine learning (ML) algorithms to parse vast biological datasets, leading to the development of predictive models for epigenetic modifications and providing a framework for understanding complex regulatory mechanisms, such as those governing cell identity genes. It supports innovative platforms like CEFCIG for high-accuracy predictions and tools like GridGO for tailored ChIP-Seq analysis, which are vital for deciphering the epigenetic landscape. The review also casts a vision on the prospects of AI and ML in oncology, particularly in the personalization of cancer therapy, including early diagnostics and treatment optimization for diseases like head and neck and colorectal cancers by harnessing computational methods, AI advancements and integrated clinical data for a transformative impact on healthcare outcomes.

Deep Learning in Protein Structural Modeling and Design

Deep learning is catalyzing a scientific revolution fueled by big data, accessible toolkits, and powerful computational resources, impacting many fields, including protein structural modeling. Protein structural modeling, such as predicting structure from amino acid sequence and evolutionary information, designing proteins toward desirable functionality, or predicting properties or behavior of a protein, is critical to understand and engineer biological systems at the molecular level. In this review, we summarize the recent advances in applying deep learning techniques to tackle problems in protein structural modeling and design. We dissect the emerging approaches using deep learning techniques for protein structural modeling and discuss advances and challenges that must be addressed. We argue for the central importance of structure, following the "sequence → structure → function" paradigm. This review is directed to help both computational biologists to gain familiarity with the deep learning methods applied in protein modeling, and computer scientists to gain perspective on the biologically meaningful problems that may benefit from deep learning techniques.

Engineering protein fragments via evolutionary and protein-protein interaction algorithms: de novo design of peptide inhibitors for FO F1 -ATP synthase

Enzyme subunit interfaces have remarkable potential in drug design as both target and scaffold for their own inhibitors. We show an evolution-driven strategy for the de novo design of peptide inhibitors targeting interfaces of the Escherichia coli FoF1-ATP synthase as a case study. The evolutionary algorithm ROSE was applied to generate diversity-oriented peptide libraries by engineering peptide fragments from ATP synthase interfaces. The resulting peptides were scored with PPI-Detect, a sequence-based predictor of protein-protein interactions. Two selected peptides were confirmed by in vitro inhibition and binding tests. The proposed methodology can be widely applied to design peptides targeting relevant interfaces of enzymatic complexes.

Computer-Aided Design of Antimicrobial Peptides: Are We Generating Effective Drug Candidates?

Antimicrobial peptides (AMPs), especially antibacterial peptides, have been widely investigated as potential alternatives to antibiotic-based therapies. Indeed, naturally occurring and synthetic AMPs have shown promising results against a series of clinically relevant bacteria. Even so, this class of antimicrobials has continuously failed clinical trials at some point, highlighting the importance of AMP optimization. In this context, the computer-aided design of AMPs has put together crucial information on chemical parameters and bioactivities in AMP sequences, thus providing modes of prediction to evaluate the antibacterial potential of a candidate sequence before synthesis. Quantitative structure-activity relationship (QSAR) computational models, for instance, have greatly contributed to AMP sequence optimization aimed at improved biological activities. In addition to machine-learning methods, the de novo design, linguistic model, pattern insertion methods, and genetic algorithms, have shown the potential to boost the automated design of AMPs. However, how successful have these approaches been in generating effective antibacterial drug candidates? Bearing this in mind, this review will focus on the main computational strategies that have generated AMPs with promising activities against pathogenic bacteria, as well as anti-infective potential in different animal models, including sepsis and cutaneous infections. Moreover, we will point out recent studies on the computer-aided design of antibiofilm peptides. As expected from automated design strategies, diverse candidate sequences with different structural arrangements have been generated and deposited in databases. We will, therefore, also discuss the structural diversity that has been engendered.

In silico design and optimization of selective membranolytic anticancer peptides

Membranolytic anticancer peptides represent a potential strategy in the fight against cancer. However, our understanding of the underlying structure-activity relationships and the mechanisms driving their cell selectivity is still limited. We developed a computational approach as a step towards the rational design of potent and selective anticancer peptides. This machine learning model distinguishes between peptides with and without anticancer activity. This classifier was experimentally validated by synthesizing and testing a selection of 12 computationally generated peptides. In total, 83% of these predictions were correct. We then utilized an evolutionary molecular design algorithm to improve the peptide selectivity for cancer cells. This simulated molecular evolution process led to a five-fold selectivity increase with regard to human dermal microvascular endothelial cells and more than ten-fold improvement towards human erythrocytes. The results of the present study advocate for the applicability of machine learning models and evolutionary algorithms to design and optimize novel synthetic anticancer peptides with reduced hemolytic liability and increased cell-type selectivity.

Autonomous Molecular Design: Then and Now

The success of deep machine learning in processing of large amounts of data, for example, in image or voice recognition and generation, raises the possibilities that these tools can also be applied for solving complex problems in materials science. In this forum article, we focus on molecular design that aims to answer the question on how we can predict and synthesize molecules with tailored physical, chemical, or biological properties. A potential answer to this question could be found by using intelligent systems that integrate physical models and computational machine learning techniques with automated synthesis and characterization tools. Such systems learn through every single experiment in an analogy to a human scientific expert. While the general idea of an autonomous system for molecular synthesis and characterization has been around for a while, its implementations for the materials sciences are sparse. Here we provide an overview of the developments in chemistry automation and the applications of machine learning techniques in the chemical and pharmaceutical industries with a focus on the novel capabilities that deep learning brings in.

Automating drug discovery

Small-molecule drug discovery can be viewed as a challenging multidimensional problem in which various characteristics of compounds - including efficacy, pharmacokinetics and safety - need to be optimized in parallel to provide drug candidates. Recent advances in areas such as microfluidics-assisted chemical synthesis and biological testing, as well as artificial intelligence systems that improve a design hypothesis through feedback analysis, are now providing a basis for the introduction of greater automation into aspects of this process. This could potentially accelerate time frames for compound discovery and optimization and enable more effective searches of chemical space. However, such approaches also raise considerable conceptual, technical and organizational challenges, as well as scepticism about the current hype around them. This article aims to identify the approaches and technologies that could be implemented robustly by medicinal chemists in the near future and to critically analyse the opportunities and challenges for their more widespread application.

Peptide-Membrane Interaction between Targeting and Lysis

Certain cationic peptides interact with biological membranes. These often-complex interactions can result in peptide targeting to the membrane, or in membrane permeation, rupture, and cell lysis. We investigated the relationship between the structural features of membrane-active peptides and these effects, to better understand these processes. To this end, we employed a computational method for morphing a membranolytic antimicrobial peptide into a nonmembranolytic mitochondrial targeting peptide by "directed simulated evolution." The results obtained demonstrate that superficially subtle sequence modifications can strongly affect the peptides' membranolytic and membrane-targeting abilities. Spectroscopic and computational analyses suggest that N- and C-terminal structural flexibility plays a crucial role in determining the mode of peptide-membrane interaction.

Computational Methods Applied to Rational Drug Design

Due to the synergic relationship between medical chemistry, bioinformatics and molecular simulation, the development of new accurate computational tools for small molecules drug design has been rising over the last years. The main result is the increased number of publications where computational techniques such as molecular docking, de novo design as well as virtual screening have been used to estimate the binding mode, site and energy of novel small molecules. In this work I review some tools, which enable the study of biological systems at the atomistic level, providing relevant information and thereby, enhancing the process of rational drug design.

Self Organizing Map-Based Classification of Cathepsin k and S Inhibitors with Different Selectivity Profiles Using Different Structural Molecular Fingerprints: Design and Application for Discovery of Novel Hits

The main step in a successful drug discovery pipeline is the identification of small potent compounds that selectively bind to the target of interest with high affinity. However, there is still a shortage of efficient and accurate computational methods with powerful capability to study and hence predict compound selectivity properties. In this work, we propose an affordable machine learning method to perform compound selectivity classification and prediction. For this purpose, we have collected compounds with reported activity and built a selectivity database formed of 153 cathepsin K and S inhibitors that are considered of medicinal interest. This database has three compound sets, two K/S and S/K selective ones and one non-selective KS one. We have subjected this database to the selectivity classification tool ‘Emergent Self-Organizing Maps’ for exploring its capability to differentiate selective cathepsin inhibitors for one target over the other. The method exhibited good clustering performance for selective ligands with high accuracy (up to 100 %). Among the possibilites, BAPs and MACCS molecular structural fingerprints were used for such a classification. The results exhibited the ability of the method for structure-selectivity relationship interpretation and selectivity markers were identified for the design of further novel inhibitors with high activity and target selectivity.

Piloting the membranolytic activities of peptides with a self-organizing map

Antimicrobial peptides (AMPs) show remarkable selectivity toward lipid membranes and possess promising antibiotic potential. Their modes of action are diverse and not fully understood, and innovative peptide design strategies are needed to generate AMPs with improved properties. We present a de novo peptide design approach that resulted in new AMPs possessing low-nanomolar membranolytic activities. Thermal analysis revealed an entropy-driven mechanism of action. The study demonstrates sustained potential of advanced computational methods for designing peptides with the desired activity.

Combinatorial chemistry by ant colony optimization

Background: Prioritizing building blocks for combinatorial medicinal chemistry represents an optimization task. We present the application of an artificial ant colony algorithm to combinatorial molecular design (Molecular Ant Algorithm [MAntA]). Results: In a retrospective evaluation, the ant algorithm performed favorably compared with other stochastic optimization methods. Application of MAntA to peptide design resulted in new octapeptides exhibiting substantial binding to mouse MHC-I (H-2Kb). In a second study, MAntA generated a new functional factor Xa inhibitor by Ugi-type three-component reaction. Conclusion: This proof-of-concept study validates artificial ant systems as innovative computational tools for efficient building block prioritization in combinatorial chemistry. Focused activity-enriched compound collections are obtained without the need for exhaustive product enumeration.

Antimicrobial peptides

The rapid increase in drug-resistant infections has presented a serious challenge to antimicrobial therapies. The failure of the most potent antibiotics to kill “superbugs” emphasizes the urgent need to develop other control agents. Here we review the history and new development of antimicrobial peptides (AMPs), a growing class of natural and synthetic peptides with a wide spectrum of targets including viruses, bacteria, fungi, and parasites. We summarize the major types of AMPs, their modes of action, and the common mechanisms of AMP resistance. In addition, we discuss the principles for designing effective AMPs and the potential of using AMPs to control biofilms (multicellular structures of bacteria embedded in extracellular matrixes) and persister cells (dormant phenotypic variants of bacterial cells that are highly tolerant to antibiotics).

Molecular evolution of peptide ligands with custom-tailored characteristics for targeting of glycostructures

As an advanced approach to identify suitable targeting molecules required for various diagnostic and therapeutic interventions, we developed a procedure to devise peptides with customizable features by an iterative computer-assisted optimization strategy. An evolutionary algorithm was utilized to breed peptides in silico and the “fitness” of peptides was determined in an appropriate laboratory in vitro assay. The influence of different evolutional parameters and mechanisms such as mutation rate, crossover probability, gaussian variation and fitness value scaling on the course of this artificial evolutional process was investigated. As a proof of concept peptidic ligands for a model target molecule, the cell surface glycolipid ganglioside G_M1, were identified. Consensus sequences describing local fitness optima were reached from diverse sets of L- and proteolytically stable D lead peptides. Ten rounds of evolutional optimization encompassing a total of just 4400 peptides lead to an increase in affinity of the peptides towards fluorescently labeled ganglioside G_M1 by a factor of 100 for L- and 400 for D-peptides.

A clever identification procedure is crucial when peptidic ligands for diagnostic and therapeutic techniques such as in vivo imaging or drug targeting are to be developed. Here, we present a propitious and versatile approach for the discovery of peptide sequences with custom features that is based on an iterative computer-assisted optimization process. The methodology smartly combines in silico evolution with in vitro testing to quickly obtain promising peptide ligand candidates with desired properties. To validate our method in a proof of concept we tried to identify peptide sequences that can bind to a glycosidic cell membrane component. We applied the evolution process by starting out with a small population of peptide lead sequences and achieved a constant increase in affinity between the peptide candidates and their target molecule with each generation. After 10 rounds and a total number of only 4400 peptides synthesized and tested, a more than 100fold improvement in target recognition could be achieved. Since all kinds of building blocks useable in chemical solid phase peptide synthesis can in principle be employed in this evolutionary optimization process, our method should prove a most versatile approach for the optimization of peptides, peptoids and peptomers towards a preset functionality.

Molecular evolution of a peptide GPCR ligand driven by artificial neural networks

Peptide ligands of G protein-coupled receptors constitute valuable natural lead structures for the development of highly selective drugs and high-affinity tools to probe ligand-receptor interaction. Currently, pharmacological and metabolic modification of natural peptides involves either an iterative trial-and-error process based on structure-activity relationships or screening of peptide libraries that contain many structural variants of the native molecule. Here, we present a novel neural network architecture for the improvement of metabolic stability without loss of bioactivity. In this approach the peptide sequence determines the topology of the neural network and each cell corresponds one-to-one to a single amino acid of the peptide chain. Using a training set, the learning algorithm calculated weights for each cell. The resulting network calculated the fitness function in a genetic algorithm to explore the virtual space of all possible peptides. The network training was based on gradient descent techniques which rely on the efficient calculation of the gradient by back-propagation. After three consecutive cycles of sequence design by the neural network, peptide synthesis and bioassay this new approach yielded a ligand with 70fold higher metabolic stability compared to the wild type peptide without loss of the subnanomolar activity in the biological assay. Combining specialized neural networks with an exploration of the combinatorial amino acid sequence space by genetic algorithms represents a novel rational strategy for peptide design and optimization.

Designing antimicrobial peptides: form follows function

Multidrug-resistant bacteria are a severe threat to public health. Conventional antibiotics are becoming increasingly ineffective as a result of resistance, and it is imperative to find new antibacterial strategies. Natural antimicrobials, known as host defence peptides or antimicrobial peptides, defend host organisms against microbes but most have modest direct antibiotic activity. Enhanced variants have been developed using straightforward design and optimization strategies and are being tested clinically. Here, we describe advanced computer-assisted design strategies that address the difficult problem of relating primary sequence to peptide structure, and are delivering more potent, cost-effective, broad-spectrum peptides as potential next-generation antibiotics.

Adaptive combinatorial design of focused compound libraries

Low-throughput screening for bioactive substances often represents the only way to discover new ligands of a drug target. This limits the number of compounds that can be tested for bioactivity. In such a situation, the design of small, focused compound libraries provides an alternative to the concept of large, maximally diverse screening collections. We present the technique of "adaptive" compound library design, which implements a simulated evolutionary process. Compound assembly and determination of bioactivity can be performed using computer-based methods (virtual screening), or in the laboratory. We show that there exists an optimal combination of the size of a screening library and the number of iterative screening rounds with the aim to keep experimental efforts at a minimum.

The therapeutic potential of immune therapy in idiopathic dilated cardiomyopathy

Idiopathic dilated cardiomyopathy (DCM) is a heart muscle disease of unknown origin and the second most common reason for heart transplantation. Genetic factors, viral persistence and the presence of an autoimmune response against myocardial epitopes are believed to play a major pathogenic role in idiopathic DCM. Pathogenic circulating autoantibodies against several cardiac proteins have been detected in sera from idiopathic DCM patients. Accordingly, suppression of autoreactive components of the immune system has been discussed as a prospective therapeutic implement in idiopathic DCM management. Removal of pathophysiologic active autoantibodies by immunoadsorption and subsequent immunoglobulin infusion induces acute and long-term beneficial effects such as improved cardiovascular function and reduced morbidity. Understanding of the amendment in various components of the immune system induced by immunoadsorption therapy may help in elucidating the underlying pathophysiologic mechanisms of the disease. This approach may reveal markers of prognostic value, and new therapeutic approaches could be established.

ENPDA: an evolutionary structure-based de novo peptide design algorithm

One of the goals of computational chemists is to automate the de novo design of bioactive molecules. Despite significant advances in computational approaches to ligand design and binding energy evaluation, novel procedures for ligand design are required. Evolutionary computation provides a new approach to this design endeavor. We propose an evolutionary tool for de novo peptide design, based on the evaluation of energies for peptide binding to a user-defined protein surface patch. Special emphasis has been placed on the evaluation of the proposed peptides, leading to two different evaluation heuristics. The software developed was successfully tested on the design of ligands for the proteins prolyl oligopeptidase, p53, and DNA gyrase.

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.

Application of the “Codon-shuffling” Method

Library-based methods of non-rational and part-rational designed de novo peptides are worthy beacons in the search for bioactive peptides and proteins of medicinal importance. In this report, we have used a recently developed directed evolution method called "codon shuffling" for the synthesis and selection of bioactive proteins. The selection of such proteins was based on the creation of an inducible library of "codon-shuffled" genes that are constructed from the ligation-based assembly of judiciously designed hexamer DNA duplexes called dicodons. Upon induction with isopropyl 1-thio-beta-D-galactopyranoside, some library members were found to express dicodon-incorporated proteins. Because of this, the host cells, in our case Escherichia coli, were unable to grow any further. The bactereostatic/lytic nature of the dicodon proteins was monitored by growth curves as well as by zone clearance studies. Transmission electron microscopy of the affected cells illustrated the extent of cell damage. The proteins themselves were overexpressed as fusion partners and subsequently purified to homogeneity. One such purified protein was found to strongly bind heparin, an indication that the interaction of the de novo proteins may be with the nucleic acids of the host cell, much like many of the naturally occurring antibacterial peptides, e.g. Buforin. Therefore, our approach may help in generating a multitude of finely tuned antibacterial proteins that can potentially be regarded as lead compounds once the method is extended to pathogenic hosts, such as Mycobacteria, for example.

Directed molecular evolution by machine learning and the influence of nonlinear interactions

Alternative search strategies for the directed evolution of proteins are presented and compared with each other. In particular, two different machine learning strategies based on partial least-squares regression are developed: the first contains only linear terms that represent a given residue's independent contribution to fitness, the second contains additional nonlinear terms to account for potential epistatic coupling between residues. The nonlinear modeling strategy is further divided into two types, one that contains all possible nonlinear terms and another that makes use of a genetic algorithm to select a subset of important interaction terms. The performance of each modeling type as a function of training set size is analysed. Simulated molecular evolution on a synthetic protein landscape shows the use of machine learning techniques to guide library design can be a powerful addition to library generation methods such as DNA shuffling.

Putting engineering back into protein engineering: bioinformatic approaches to catalyst design

Complex multivariate engineering problems are commonplace and not unique to protein engineering. Mathematical and data-mining tools developed in other fields of engineering have now been applied to analyze sequence-activity relationships of peptides and proteins and to assist in the design of proteins and peptides with specified properties. Decreasing costs of DNA sequencing in conjunction with methods to quickly synthesize statistically representative sets of proteins allow modern heuristic statistics to be applied to protein engineering. This provides an alternative approach to expensive assays or unreliable high-throughput surrogate screens.

Virtual screening and fast automated docking methods

Recent advances in high-throughput protein structure determination and in computational chemistry have refocused attention on virtual screening and fast automated docking methods. This review provides a brief introduction to the basic ideas and outlines computational tools currently used. We also provide several examples of where virtual screening has proved successful, highlighting the usefulness of the approach.

Exploration of sequence space for protein engineering

The process of protein engineering is currently evolving towards a heuristic understanding of the sequence-function relationship. Improved DNA sequencing capacity, efficient protein function characterization and improved quality of data points in conjunction with well-established statistical tools from other industries are changing the protein engineering field. Algorithms capturing the heuristic sequence-function relationships will have a drastic impact on the field of protein engineering. In this review, several alternative approaches to quantitatively assess sequence space are discussed and the relatively few examples of wet-lab validation of statistical sequence-function characterization/correlation are described.

Design of inhibitors of Ras--Raf interaction using a computational combinatorial algorithm

Drugs that inhibit important protein-protein interactions are hard to find either by screening or rational design, at least so far. Most drugs on the market that target proteins today are therefore aimed at well-defined binding pockets in proteins. While computer-aided design is widely used to facilitate the drug discovery process for binding pockets, its application to the design of inhibitors that target the protein surface initially seems to be limited because of the increased complexity of the task. Previously, we had started to develop a computational combinatorial design approach based on the well-known 'multiple copy simultaneous search' (MCSS) procedure to tackle this problem. In order to identify sequence patterns of potential inhibitor peptides, a three-step procedure is employed: first, using MCSS, the locations of specific functional groups on the protein surface are identified; second, after constructing the peptide main chain based on the location of favorite locations of N-methylacetamide groups, functional groups corresponding to amino acid side chains are selected and connected to the main chain C(alpha) atoms; finally, the peptides generated in the second step are aligned and probabilities of amino acids at each position are calculated from the alignment scheme. Sequence patterns of potential inhibitors are determined based on the propensities of amino acids at each C(alpha) position. Here we report the optimization of inhibitor peptides using the sequence patterns determined by our method. Several short peptides derived from our prediction inhibit the Ras--Raf association in vitro in ELISA competition assays, radioassays and biosensor-based assays, demonstrating the feasibility of our approach. Consequently, our method provides an important step towards the development of novel anti-Ras agents and the structure-based design of inhibitors of protein--protein interactions.

Synthetic peptides as putative therapeutic agents in transplantation medicine: application of PEPSCAN to the identification of functional sequences in the extracellular domain of the interleukin-2 receptor beta chain (IL-2Rbeta)

A desired treatment strategy in transplantation medicine is the selective targeting of alloreactive T cells without impairing antileukemic and antiviral activities. One approach is the synthesis of peptides that interfere with the binding of interleukin-2 (IL-2) to its high affinity receptor (IL-2R). This blocks the activation and proliferation of the antigen-activated T cells and the secretion of IL-2. The latter binds to its receptor, via the extracellular domain of the IL-2Rbeta chain, while its cytoplasmic domain is required for intracellular signal transduction. In this study, the PEPSCAN method was applied in order to identify antigenic sequences (epitopes) in the extracellular domain of the IL-2Rbeta. Based on the primary amino acid (aa) sequence of the IL-2Rbeta, a total of 239 overlapping dodecapeptides, spanning the entire sequence of IL-2Rbeta, were synthesized by PEPSCAN and their immunoreactivity was tested by ELISA using monoclonal antibodies (mAbs) specific for IL-2Rbeta such as TU11, Mikbeta1, HuMikbeta1 and TU27. TU11 recognized a linear epitope located in the region 85R-Q(96). None of the 239 synthetic peptides was recognized by TU27. Mikbeta1 (and HuMikbeta1) recognized a discontinuous epitope formed by aa located in the IL-2Rbeta domains L(106) to P(148) and E(170) to A(202). Subsequently, synthetic peptides corresponding to the identified putative epitopic sequences were prepared by solid phase synthesis and their immunogenicity in vivo was assessed by raising polyclonal antibodies. Given that Mikbeta1 and HuMikbeta1 inhibit binding of IL-2 on the IL-2Rbeta, we addressed the question of whether the identified antigenic sequences serve as putative IL-2 binding domains. Synthetic peptides corresponding to these sequences were tested for their ability to compete with IL-2 for binding and, thereby, inhibit IL-2-induced proliferation of mitogen-stimulated human peripheral blood T cells. Sequences 107M-E(118) and 178Y-Q(199) probably represent functional IL-2 binding domains on IL-2Rbeta, since these synthetic peptides significantly inhibited the proliferation of activated T cells and secretion of IL-2.

A genetic algorithm for the automated generation of small organic molecules: drug design using an evolutionary algorithm

Rational drug design involves finding solutions to large combinatorial problems for which an exhaustive search is impractical. Genetic algorithms provide a novel tool for the investigation of such problems. These are a class of algorithms that mimic some of the major characteristics of Darwinian evolution. LEA has been designed in order to conceive novel small organic molecules which satisfy quantitative structure-activity relationship based rules (fitness). The fitness consists of a sum of constraints that are range properties. The algorithm takes an initial set of fragments and iteratively improves them by means of crossover and mutation operators that are related to those involved in Darwinian evolution. The basis of the algorithm, its implementation and parameterization, are described together with an application in de novo molecular design of new retinoids. The results may be promising for chemical synthesis and show that this tool may find extensive applications in de novo drug design projects.

De novo design of molecular architectures by evolutionary assembly of drug-derived building blocks

An evolutionary algorithm was developed for fragment-based de novo design of molecules (TOPAS, TOPology-Assigning System). This stochastic method aims at generating a novel molecular structure mimicking a template structure. A set of approximately 25,000 fragment structures serves as the building block supply, which were obtained by a straightforward fragmentation procedure applied to 36,000 known drugs. Eleven reaction schemes were implemented for both fragmentation and building block assembly. This combination of drug-derived building blocks and a restricted set of reaction schemes proved to be a key for the automatic development of novel, synthetically tractable structures. In a cyclic optimization process, molecular architectures were generated from a parent structure by virtual synthesis, and the best structure of a generation was selected as the parent for the subsequent TOPAS cycle. Similarity measures were used to define 'fitness', based on 2D-structural similarity or topological pharmacophore distance between the template molecule and the variants. The concept of varying library 'diversity' during a design process was consequently implemented by using adaptive variant distributions. The efficiency of the design algorithm was demonstrated for the de novo construction of potential thrombin inhibitors mimicking peptide and non-peptide template structures.

Neural networks are useful tools for drug design

A profound understanding of molecular recognition processes and the underlying molecular interaction patterns is a prerequisite for future progress and success in rational drug design. Neural networks could play an important role in guiding the Drug Discovery process through the extraction of relevant molecular features.