Swarm intelligence for chemical reaction optimization
Rémi Schlama, Joshua W. Sin, Ryan P. Burwood, Kurt Püntener, Raphael Bigler, Philippe Schwaller · Chem · 2026
We report ⍺-particle swarm optimization (PSO), a nature-inspired metaheuristic algorithm that augments canonical PSO with machine learning (ML) for highly parallel reaction optimization. Unlike black-box ML approaches that obscure decision-making processes, ⍺-PSO uses simple, physically intuitive swarm dynamics directly connected to experimental observables, enabling practitioners to understand each component driving optimization. We establish a theoretical framework for reaction landscape analysis using local Lipschitz constants to quantify reaction space "roughness," from smoothly varying reaction surfaces to landscapes with many reactivity cliffs. This analysis guides adaptive ⍺-PSO parameter selection, optimizing performance for different reaction topologies. Evaluation of ⍺-PSO across pharmaceutically relevant reaction benchmarks demonstrates competitive performance with state-of-the-art Bayesian optimization (BO) methods, whereas two prospective high-throughput experimentation (HTE) campaigns showed that ⍺-PSO identified optimal reaction conditions more rapidly than BO. Alongside our open-source ⍺-PSO implementation, we release 989 new high-quality Pd-catalyzed Buchwald-Hartwig and Suzuki reactions generated in these campaigns.