Machine Learning-Guided Development of Trialkylphosphine Ni^(I) Dimers and Applications in Site-Selective Catalysis

Abstract

Owing to the unknown correlation of a metal’s ligand and its resulting preferred speciation in terms of oxidation state, geometry, and nuclearity, a rational design of multinuclear catalysts remains challenging. With the goal to accelerate the identification of suitable ligands that form trialkylphosphine-derived dihalogen-bridged Ni(I) dimers, we herein employed an assumption-based machine learning approach. The workflow offers guidance in ligand space for a desired speciation without (or only minimal) prior experimental data points. We experimentally verified the predictions and synthesized numerous novel Ni(I) dimers as well as explored their potential in catalysis. We demonstrate C–I selective arylations of polyhalogenated arenes bearing competing C–Br and C–Cl sites in under 5 min at room temperature using 0.2 mol % of the newly developed dimer, [Ni(I)(μ-Br)­PAd2(n-Bu)]2, which is so far unmet with alternative dinuclear or mononuclear Ni or Pd catalysts.

Synthesis of Nickel(I)–Bromide Complexes via Oxidation and Ligand Displacement: Evaluation of Ligand Effects on Speciation and Reactivity

“…Notably, small monophosphine ligands have been known to form monomeric (PR 3 ) n Ni I X (n = 2 or 3) species via comproportionation and oxidative addition, 26,27 but dimeric [(PR 3 )Ni I X] 2 complexes bearing larger phosphine congeners were unknown prior to a 2022 publication from our group and Schoenebeck's report concurrent to this work. 28,29 This is in stark contrast to the precedent for Pd I , for which complexes of the type [(PR 3 )-Pd I X] 2 are long-known 30 Schoenebeck and co-workers. 31,32 As the precedent for bisligated (PR 3 ) 2 Ni I species has shown that such species are capable of catalytic cycle (re)entry, 13,18−20 the structure and speciation of monoligated (PR 3 )Ni I complexes warrants further investigation.…”

Chemoinformatic Catalyst Selection Methods for the Optimization of Copper–Bis(oxazoline)-Mediated, Asymmetric, Vinylogous Mukaiyama Aldol Reactions

“…LOCO validation ensures that every prediction is out-of-sample with respect to catalyst identity, simulating the scenario when predicting out-of-sample in silico structures. Attempts at featurizing substrate combinations and products with CAGB ASO and average electronic indicator field (AEIF) descriptors, sterimol descriptors, and various DFT descriptors did not result in more predictive models compared to simple one-hot encoded substrate descriptors. Feature selection was employed to produce models with decent predictive performance ( R 2 = 0.72) and absolute errors (0.17 kcal/mol, Figure ).…”

“…Sigman et al have identified a representative, mostly achiral ligand subset in this database and recently disclosed chiral bisphosphine ligand QSSR models for Hayashi–Heck reactions . Data-driven workflows with a general form of (1) featurizing broad chemical spaces, (2) selecting diverse representative subsets, (3) acquiring empirical data in some reaction manifold, and (4) constructing QSSR models for the potential prediction of new optimal catalysts are becoming an increasingly popular area of research. Whereas the featurization, exploration, and modeling of phosphine ligand space has received significant attention, we now apply this workflow to oxazoline-containing ligands.…”

A Photochemical Strategy for the Conversion of Nitroarenes into Rigidified Pyrrolidine Analogues

“…Typically, new bioisostere identification has taken a targeted approach, in which specific bond vectors of novel chemotypes are compared with the ones of few selected parent compounds. − We decided to utilize a different approach where the bicyclic cores were compared with a virtual database of >250 potentially interesting N-heterocycles with two vectors appended. , This strategy would allow an expansive area of chemical space to be efficiently surveyed, including unreported species. To map the chemical space to the goal of bioisostere identification, we decided to create a high-dimensional data set of relevant descriptors , where the chemical space could then be visualized and interpreted by a principal component analysis (PCA). , Such a data set would also offer the opportunity to apply unsupervised learning algorithms, to identify machine-guided relationship. − Therefore, we envisioned that applying methods such as the relatively straightforward k -Medoids clustering − would provide an unbiased insight into bioisostere identification.…”