Virtual screening in the search for new drug-like compounds plays a key role in the field of computer-aided drug design. In library optimization, the task is to search a subset of reagents to have the optimal set of products. For many cases there are too many combinatorial products to take an exhaustive search under the combinatorial constraints. Most often, genetic algorithm is used to find the high-scoring molecules in the total combinatorial space but genetic algorithm is slow and unstable due to the randomness in the procedure itself. In this study, the starting point for combinatorial library optimization is the calculated properties of each compound in terms of the drug likeness. Instead of using traditional genetic algorithm for random search, we design machine learning-based optimization procedures to find the optimal combinatorial subset of the products. Then we compare the machine learning algorithms with genetic algorithms. Some sample data is used to demonstrate the effectiveness and performance of our algorithms.

Optimizing combinatorial libraries using machine learning approaches

ACS Spring Meeting 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.