Data-Driven Insights into the Transition-Metal-Catalyzed Asymmetric Hydrogenation of Olefins

Sukriti Singh, José Miguel Hernández-LobatoThe Journal of Organic Chemistry 2024 · DOI

TL;DR

This paper leverages a large literature dataset to perform a data-driven analysis of transition-metal-catalyzed asymmetric hydrogenation (AHO) of olefins, focusing on iridium, rhodium, and cobalt. It reveals significant biases in reported reactions, highlighting under-explored areas and providing valuable insights for designing more effective machine-learning models for reaction optimization and catalyst discovery.

What Problem Does This Paper Address?

Asymmetric hydrogenation of olefins (AHO) is a cornerstone reaction in organic synthesis, enabling the creation of chiral molecules crucial for pharmaceuticals and fine chemicals. While iridium and rhodium catalysts have been extensively studied, the use of earth-abundant cobalt is a more recent development. The field has generated a vast amount of literature data, presenting an opportunity for data-driven approaches like machine learning (ML) to accelerate discovery and optimization. However, a significant challenge for ML models is the inherent bias in published literature. Researchers tend to report successful, high-yielding, and highly selective reactions, leading to datasets that may not represent the full chemical space or the diversity of outcomes. This bias can cause ML models to simply learn and reproduce popular trends rather than uncovering novel insights or predicting truly optimal conditions. This paper aims to address this by systematically analyzing existing AHO literature data to understand these biases and identify opportunities for more comprehensive data collection and ML model development.

Methods

The core methodology of this paper is a comprehensive data-driven analysis of a literature-mined dataset of AHO reactions. The authors employed several key techniques:

  • Data Mining and Curation: They utilized a recently published dataset containing over 12,000 AHO reactions, which includes detailed information on substrates (olefins), catalysts (metals and ligands), reaction conditions (solvent, temperature, pressure), and performance metrics (enantioselectivity, yield). For cobalt, they supplemented the dataset with manually extracted data from recent literature to address its underrepresentation.
  • Statistical Analysis: They performed detailed statistical analyses to categorize and quantify the usage of different olefin types, ligand classes, and reaction conditions across iridium, rhodium, and cobalt catalysts. This involved calculating frequencies and distributions.
  • Chemical Space Visualization: To understand the relationships and overlaps between different components of the AHO reaction, they employed dimensionality reduction techniques, specifically Uniform Manifold Approximation and Projection (UMAP). This allowed them to visualize the chemical space occupied by olefins and ligands, revealing clusters and areas of sparse exploration.
  • Olefin-Ligand Combination Analysis: A crucial aspect was analyzing the combinations of specific olefin types with specific ligand classes. This was visualized using scatter plots where circle size indicated reaction count and color represented median enantioselectivity, highlighting popular and under-explored pairings.

Datasets

The primary dataset used in this study is a literature-mined collection of transition-metal-catalyzed asymmetric hydrogenation of olefins (AHO). This dataset, referred to as "asymcatml.net," contains approximately 12,619 reactions, encompassing 2,754 unique olefin substrates and 1,686 ligands. For each reaction, the dataset includes SMILES strings for substrates, ligands, solvents, and products, along with experimental variables like pressure, temperature, and catalyst loading. Performance data such as enantiomeric excess (ee), yield, and conversion are also reported. The authors note that the data for cobalt-catalyzed AHO reactions was limited (1% of the total), and they manually extracted additional data from recent publications to provide a more comprehensive analysis for this metal.

Key Insights and Ideas

  • Literature Bias is Pervasive: The study strongly confirms that the published literature on AHO is heavily biased towards high enantioselectivity and yield. This means that datasets derived solely from published successes may not accurately reflect the true scope of what is achievable or the challenges encountered.
  • Cobalt as a Promising, Under-Explored Area: While iridium and rhodium dominate the literature, cobalt catalysis is emerging. The analysis suggests that cobalt catalysts can offer unique reactivity and are a cost-effective alternative, but the exploration of its ligand and substrate space is significantly less developed compared to its noble metal counterparts.
  • Distinct Catalytic Niches: Despite being in the same periodic group, Ir, Rh, and Co catalysts exhibit distinct preferences for olefin substrates and ligand types. For instance, Ir favors unfunctionalized olefins with P,N ligands, while Rh leans towards functionalized olefins with phosphine-based ligands.
  • Chemical Space Visualization Reveals Gaps: UMAP plots effectively illustrate that the chemical space of both olefins and ligands is not uniformly explored. This visualization highlights opportunities for researchers to target specific under-explored regions to discover new catalysts or reaction conditions.
  • Olefin-Ligand Combinations are Sparse: The analysis of olefin-ligand combinations reveals that only a limited number of pairings are frequently reported. This sparsity suggests that many potentially effective combinations remain undiscovered or unreported.

Key Results

  • Catalyst Distribution: Iridium and rhodium account for the vast majority of reported AHO reactions (53.1% and 37.9%, respectively), with cobalt representing a mere 1% in the initial dataset.
  • Substrate Preferences: Trisubstituted olefins are the most common substrate type overall. However, specific preferences emerge: Ir-catalyzed reactions often involve aryl/alkyl-substituted alkenes, while Rh-catalyzed reactions frequently utilize α,β-unsaturated carbonyls and enamides. Cobalt shows a mix, with aryl/alkyl-substituted alkenes dominating for disubstituted olefins.
  • Ligand Dominance: P,N ligands are dominant for Ir catalysis. For Rh, monodentate and bidentate phosphorus ligands (especially phosphoramidites and bisphosphines) are prevalent. Cobalt's ligand landscape is smaller but resembles Rh's, with bisphosphines being popular.
  • Reaction Conditions: Dichloromethane (DCM) is the overwhelmingly preferred solvent for Ir and Rh catalysis (around 80% and 40%, respectively). Cobalt shows a greater preference for alcohol solvents and a notable use of nonpolar solvents like toluene. Room temperature is common for Ir and Rh, while cobalt reactions are more frequently conducted at elevated temperatures.
  • Pressure Trends: Ir catalysts are often used at higher pressures (49-50 atm), while Rh reactions tend to occur at lower pressures (<50 atm). Cobalt shows a more mixed pressure profile.

Limitations and Future Work

  • Data Bias: The most significant limitation acknowledged is the inherent bias in the literature data towards successful outcomes. This means that the analysis might not fully capture the challenges or the full range of reactivity. The authors emphasize the need for reporting failed or low-yielding reactions to create more balanced datasets.
  • Limited Cobalt Data: While efforts were made to supplement the cobalt data, it remains significantly less extensive than for Ir and Rh. This limits the depth of conclusions that can be drawn for cobalt catalysis.
  • Future Directions: The authors suggest that understanding these literature biases is crucial for developing more effective machine-learning models. They advocate for exploring a wider range of substrates, ligands, and reaction conditions, and for prioritizing the reporting of diverse experimental outcomes. This will enable ML models to move beyond simply learning literature trends and to truly assist in the discovery of novel catalytic systems and optimized reaction pathways.

Summary updated

Abstract

The transition-metal-catalyzed asymmetric hydrogenation of olefins is one of the key transformations with great utility in various industrial applications. The field has been dominated by the use of noble metal catalysts, such as iridium and rhodium. The reactions with the earth-abundant cobalt metal have increased only in recent years. In this work, we analyze the large amount of literature data available on iridium- and rhodium-catalyzed asymmetric hydrogenation. The limited data on reactions using Co catalysts are then examined in the context of Ir and Rh to obtain a better understanding of the reactivity pattern. A detailed data-driven study of the types of olefins, ligands, and reaction conditions such as solvent, temperature, and pressure is carried out. Our analysis provides an understanding of the literature trends and demonstrates that only a few olefin-ligand combinations or reaction conditions are frequently used. The knowledge of this bias in the literature data toward a certain group of substrates or reaction conditions can be useful for practitioners to design new reaction data sets that are suitable to obtain meaningful predictions from machine-learning models.
BibTeX
@article{singh2024data,
  author = {Sukriti Singh and José Miguel Hernández-Lobato},
  title = {{Data-Driven Insights into the Transition-Metal-Catalyzed Asymmetric Hydrogenation of Olefins}},
  journal = {The Journal of Organic Chemistry},
  year = {2024},
  doi = {10.1021/acs.joc.4c01396},
}
Figure