Evolutionary-Algorithm-Based Strategy for Computer-Assisted Structure Elucidation

Abstract

An evolutionary algorithm (EA) using a graph-based data structure to explore the molecular constitution space is presented. The EA implementation proves to be a promising alternative to deterministic approaches to the problem of computer-assisted structure elucidation (CASE). While not relying on any external database, the EA-guided CASE program SENECA is able to find correct solutions within calculation times comparable to that of other CASE expert systems. The implementation presented here significantly expands the size limit of constitutional optimization problems treatable with evolutionary algorithms by introducing novel efficient graph-based genetic operators. The new EA-based search strategy is discussed including the underlying data structures, component design, parameter optimization, and evolution process control. Typical structure elucidation examples are given to demonstrate the algorithm's performance.

Building blocks for automated elucidation of metabolites: natural product-likeness for candidate ranking

“…The overall scheme implemented by us for the optimization problem is similar to the scheme reported in [9], which has both mutation and recombination operators. In our scheme, we used only the mutation operator because it offered the desired results and was easier to configure.…”

Automated Interpretation of NMR Spectra for Small Organic Molecules in Solution

“…32,33 Likeliness is judged from diverse criteria, such as the matching between predicted and experimental chemical shifts, the length of the coupling paths in accordance with observed 2-D NMR correlations, the presence of atom groups in symmetrical positions, the presence or absence of particular substructures and the conformity to Bredt's rule (no double bond at bridgehead in strained bicyclic ring systems). 34 An NP likeness criterion has been defined in order to guide the structure generation toward NP structures when this option makes sense.…”

CASE 2D NMR-based Expert Systems

“…A huge number of isomers corresponding to molecules containing 30-100 skeletal atoms resulted in some researchers [60][61][62] declaring that, in principle, deterministic systems would not be able to analyze the entire space of potential isomers and therefore fail in the elucidation of large structures. As an alternative, they suggested stochastic algorithms for structure generation, specifically simulated annealing 60,61 and genetic algorithms.…”

“…As an alternative, they suggested stochastic algorithms for structure generation, specifically simulated annealing 60,61 and genetic algorithms. 62,63 The number of skeletal atoms in a molecule that could be identified by deterministic systems was limited by these authors 61,62,64 to 30 or less. Examples of large molecules that were successfully elucidated and reported, but above the limit of 30 atoms, were considered to be exceptions.…”