Alchemical space exploration in drug design via stochastic expanding boundary optimization search

In this work, we propose a stochastic algorithm for the discovery of organic molecules with specific properties by steering alchemical changes, allowing the sampling of chemical space for potential drug candidates. The algorithm employs a tree-like local search capable of overcoming entrapment in local "basins" and iteratively sampling adjacent basins, creating a molecular ensemble with specific properties. The method consists of iterative stochastic "moves", which may alter the chemical constitution of a potential drug leading towards molecules with desirable properties. Implementation of this kind of procedure necessitates the Simplified Molecular Input Line Entry System (SMILES) molecular representation. The method has been designed especially for searches in targeted discontinuous spaces, where local optimal solutions are expected to be well separated and hard to traverse. We demonstrate our method in three applications: (i) generating ensembles of organic molecules with preset octanol-water partition coefficients, (ii) identifying molecules with specific light absorption properties via quantum mechanical calculations, and (iii) optimizing drug binding free energies to protein targets using molecular docking.