Generation of Target-Selective Drug Candidate Structures Using Molecular Evolutionary Algorithmwith SVM Classifiers

Abstract

In a previous study, we reported a molecular evolutionary approach for generating chemical structures. It involved a computational experiment for reproducing a target chemical structure from a seed structure by using a fitness based on the structural similarity. In this paper, we describe a method of molecular evolutionary computation using support vector machine (SVM) classifiers for generating drug-like candidate structures with specific activity. The method is based on evolutionary operations such as crossover, mutation, and selections similar to the previous study; however, the fitness of each structure was evaluated using the SVM classifiers. We performed molecular evolutionary computation using the SVM classifiers in order to generate candidate chemical structures for antihypertensive drugs of two different therapeutic classes of angiotensin converting enzyme (ACE) and neutral endopeptidase (NEP). A computer experiment to generate the ACE-selective candidates showed that evolutionary computation could favorably increase the fitness for ACE as the alternation of generations proceeded. Another computer experiment for the NEP-selective candidates also yielded a favorable result.

Bayesian molecular design with a chemical language model

“…This issue is common to the graph enumeration. To avoid the emergence of unfavorable structures, exclusion rules were employed in some studies, particularly those aimed at the design of drug-like molecules [ 16 , 17 ]. However, such rules might be incomprehensive, and it is impractical to establish a general rule of chemically favorable structures.…”

Automated Design of Realistic Organometallic Molecules from Fragments

“…[55][56][57][58][59][60][61]38,62 However, the major drawback of atom-based builders is that most of the combinations of atoms do not represent stable molecules or are not synthetically accessible, thus strategies to control the assembling of atoms have to be included. 63 The problem is even worse for transition metal compounds, due to the high coordination number and extended list of bond types. 30,64 In general, automated molecular builders need to balance novelty and confined chemical space.…”

“…Accordingly, graph-based representations such as connection tables and SMILES have been widely exploited as chromosome. 61,63,75,76,92 Alternatively, SMILES and 2D graphs have been used to represent only the molecular structure of the building blocks, i.e., the content of the vertices, while a dedicated data structure defined the surrounding graph-like chromosome. 30,73,91,118,119 The analogous strategy with 3D representation has been used, 58,66,74,88,[120][121][122] but mostly in connection to fragment-based drug discovery, 123 which should not be confused with the fragment-based building strategy introduced above, although the latter is usually implemented as part of the former.…”

<Review> A Mini-review on Chemoinformatics Approaches for Drug Discovery

“…On the other hand, many ligand-based approaches use a similarity-based scoring function, which calculates molecular similarity between the designed molecule and the known active molecule. Figure 3 shows the general scheme of evolutionary algorithms reported in the literature [73] [76]. Evolutionary algorithms are based on concepts derived from biological evolution, including reproduction, mutation, crossover, and selection and new molecules are designed by repeated application of evolutionary operations.…”

“…Generally, fragment-based mutations are better considered to generate drug-like structures. Douguet et al [68] and Kawai et al [76] have respectively investigated atom-based mutation and both of the result yielded unfavorable structures such as invalid hetero-hetero atom bonds. Recently, Kawai et al proposed a fragment-based mutation for generating new drug-like molecules [77].…”