Leveraging Data Science to Elucidate Ligand Features for Pd-Catalyzed Enantioretentive N‑Arylations of Cyclic α‑Substituted Amines in Aqueous Media

The combination of high-throughput experimentation (HTE) and data science offers a promising solution for the optimization of challenging chemical reactions, although current data science approaches to deal with highly skewed data sets resulting from typical HTE campaigns are limited. One such attractive yet challenging reaction is the preparation of chiral tertiary anilines via palladium-catalyzed N-arylation of enantioenriched, α-substituted secondary amines. While enantioenriched tertiary anilines are highly valuable motifs in medicinal chemistry, their synthesis remains challenging to accomplish in high yields and enantiospecificity. Herein, we disclose a method for the enantioretentive N-arylation of cyclic secondary amines. After an extensive HTE campaign, a novel phosphorinane ligand for palladium, ‘NiniPhos,’ was found to demonstrate high yields (up to 96%) and enantiospecificity (96 to >99% es) across a range of amines with diverse aryl halides under aqueous conditions. Despite screening more than 120 ligands, only 31 demonstrated catalytic activity (>10% yield), resulting in a highly skewed data set. By applying data science-driven hotspot analysis and machine learning (linear support vector machines), the structural features of ligands leading to the activity cliff were identified, pinpointing di-tert-butylphosphines and phosphorinanes as privileged ligand classes in this reaction. This workflow offers a generalizable strategy for extracting mechanistic insights from skewed data sets, which can be leveraged for reaction design and optimization.