Multi-Agent Architecture for Parameter Configuration in Computational Chemistry

Abstract

Chemists often rely on computational prediction techniques to optimise laboratory experiments. However, selecting and configuring these techniques can be a daunting task for those without a background in computer science. A potential solution arises when chemists collaborate with computer scientists to define and implement the prediction task. Based on this collaboration, we propose a novel architecture that reflects key aspects of this interaction: the expertise of computer scientists, their commitment to addressing the prediction task, and the discussions that facilitate consensus. Our proposed architecture enables chemists to generate solution proposals across the various stages of the prediction process, leveraging consensus-based decisions made by intelligent agents and incorporating a lazy learning approach.