A knowledge‐based approach for screening chemical structures withinde novomolecular evolution

Abstract

Despite the advantage of employing evolutionary algorithms (EAs) for de novo creation of novel molecular structures with optimal properties, the approach is hampered by sampling chemically undesirable structures. Such structures are undesirable for different reasons, such as a critical structural pattern may be ignored or too many rotational degrees of freedom exist for conformational search. A new method is presented which creates a user-defined structure filter, here referred to as the bias filter (BF), generated from a set of molecules representing typical structure types that are allowed to evolve in the de novo process. The BF can be seen as constraining the chemical structure space according to external requirements given by the user. No explicit programming of structure rules is necessary which makes the proposed method much more intuitive and user friendly than other methods commonly used in de novo applications. We have tested the proposed method in the process of evolving a set of Factor Xa inhibitors where the aim is to create molecules with optimal logP values. The de novo GeneGear system developed in our group is responsible for the corresponding evolutionary computation and bias filtering.

An Evolutionary Algorithm for de Novo Optimization of Functional Transition Metal Compounds

“…Similarly, if sufficient attention is paid to parametrization, it is likely that efficient fitness functions may be obtained from the application of computationally inexpensive methods such as ligand-field molecular mechanics (LFMM, 45 which has already been used successfully in design 77 ) and reactive force-fields (e.g., ReaxFF 46 ) and steps to integrate such fitness providers will be taken. Introduction of structural operators that take synthetic accessibility 78 into consideration, removal of undesirable structures prior to fitness calculation, 79 and dynamical updating of QSAR models by on-the-fly calculation of true fitness during EA optimizations constitute other desirable targets for future work.…”

MoleGear: A Java-Based Platform for Evolutionary De Novo Molecular Design

“…The basic scheme of EA in MoleGear is illustrated in Figure 9. A seed population including k molecules is initially constructed either from an available set of chemical structures [18] or a fragment library. All the structures are by default saturated with hydrogens and subjected to a conformational search performed by 3D builders.…”

“…By applying EA to the molecular design fields, a diverse chemical space can be searched to provide optimal, or near-optimal solutions to a wide range of objectives. So far, many studies have reported the use of EA tools for computer-aided de novo molecular design [6,15,17,18,19,20,21,22]. Evolutionary algorithms use fitness functions to determine the surviving structures, which will be used in the next generation population.…”

Inverse‐QSPR for de novo Design: A Review

“…Chu and Alsberg set a user‐custom structure filter named bias filter, which is able to restrict the generation space in accordance to a given set of relevant molecules. User inputs both a set of desired and undesired molecules.…”