Chemoton 2.0: Autonomous Exploration of Chemical Reaction Networks

Jan P. Unsleber, Stephanie A. Grimmel, Markus ReiherJournal of Chemical Theory and Computation 2022 · DOI

TL;DR

This paper introduces Chemoton 2.0, an open-source, extensible software framework for autonomously exploring vast chemical reaction networks using first-principles quantum mechanics. It presents novel algorithms for efficiently searching for elementary reaction steps and demonstrates its capability by successfully identifying a majority of known reactions and discovering many new ones across a diverse test set.

What Problem Does This Paper Address?

Understanding chemical processes hinges on mapping out their underlying reaction mechanisms, which can quickly expand into enormous, complex networks of elementary steps. Manually exploring these networks is infeasible due to the sheer volume of data and the computational cost of quantum chemical calculations. Existing automated methods often require significant user intervention or are limited to specific chemical systems. This paper tackles the challenge of creating a general, automated, and robust framework that can explore these reaction spaces efficiently and comprehensively, providing a foundation for mechanism elucidation, optimization, and more.

Methods

Chemoton 2.0 is built upon the SCINE software environment and features a modular design with a front-end, back-end, and data storage. The core of the exploration is driven by the Chemoton module, which utilizes engines (e.g., for job scheduling, compound housekeeping) and gears (specific algorithms).

Key novelties include:

  • Two new algorithms for elementary-step searches based on Newton trajectories:
    • Newton Trajectory Algorithm 1 (NT1): Pushes two predefined fragments together or pulls them apart, allowing for exploration of bond formation and dissociation.
  • Newton Trajectory Algorithm 2 (NT2): Focuses on defining specific atom pairs to form or break bonds, aiming to directly discover bond formation/dissociation events and facilitating the discovery of intramolecular rearrangements.
  • Robust data handling: A structured database stores information about structures, compounds, elementary steps, and reactions, enabling efficient querying and comparison.
  • Flexible filtering: The framework allows for the application of filters to control the exploration process and trim the combinatorial explosion of potential reactions.

Datasets

The authors assembled a diverse test set of 70 distinct sets of reactants leading to 184 reference product sets across 184 different reference reactions. This set includes:

  • The Zimmerman test set (reactions 1-27).
  • Unimolecular decomposition reactions of 3-hydroperoxypropanal (reactions 28).
  • Pericyclic reactions (reactions 29-38).
  • Small molecule test set (reactions 39-45).
  • Transition-metal-chemistry test cases (reactions 46-51, 54-60).
  • Challenging cases like HCN isomerization, quasi-barrierless reactions, radical reactions, cis-trans isomerization, and bifurcation reactions.

This dataset is made publicly available to serve as a starting point for future benchmarking.

Key Insights and Ideas

  • Autonomous exploration is achievable: Chemoton 2.0 demonstrates that large-scale, automated exploration of chemical reaction space is feasible using first-principles methods.
  • Newton trajectories are powerful: The two new Newton trajectory algorithms are effective at finding elementary steps, with NT2 showing a higher detection rate and better suitability for discovering complex rearrangements.
  • First-principles basis is crucial for generality: By relying on quantum mechanics, Chemoton avoids system-specific limitations, making it applicable to a wide range of chemical problems.
  • Trade-offs in electronic structure methods: Faster, approximate methods (like DFTB3 and GFN2-xTB) are useful for initial exploration, but more accurate, computationally expensive methods are needed for reliable results, especially for complex systems like transition metals.
  • The importance of diverse test sets: The carefully curated test set highlights the capabilities and limitations of the framework, pushing it to explore challenging chemical scenarios.

Key Results

  • High success rate for known reactions: Using the NT2 algorithm with GFN2-xTB, Chemoton found 79% of the reference reactions in the test set. With DFTB3, the success rate was 69% for reactions where parameters were available.
  • Discovery of numerous new reactions: Beyond the reference set, Chemoton discovered 44 times more reactions than were documented in the literature, many with low energy barriers, providing a more detailed picture of accessible reaction space.
  • Algorithm performance: NT2 generally outperformed NT1 in terms of detection rate, likely due to its ability to handle more complex bond modifications simultaneously.
  • Computational cost: The number of elementary step trials can be substantial, but the framework offers settings to balance computational cost with exploration breadth.

Limitations and Future Work

  • Spin state tracking: The current version has limitations in tracking and assigning spin states, which is crucial for reactions involving multiple spin states. This will be addressed in future releases.
  • Bifurcation handling: Explicit exploration and handling of reaction path bifurcations are not yet implemented, leading to statistically distributed results for such cases.
  • Symmetry breaking: For certain reactions, artificial breaking of symmetry is required for optimization, and a calibration mechanism is needed to ensure this is done minimally.
  • Exploration of barrier-free transitions: The current step trial implementation expects a transition state, which is not suitable for barrier-free reactions. A special bookkeeping procedure is needed for these.
  • Tracking identical reactant/product structures: The current definition of structures does not distinguish between identical atoms of the same element, meaning simple swaps of, e.g., hydrogen atoms are not tracked as elementary steps.
  • Future work: The authors plan to integrate microkinetic modeling, advanced electronic structure models, subsystem microsolvation, and machine learning for improved exploration. They also aim to develop a graphical user interface and a remote database.

Summary updated

Abstract

Fueled by advances in hardware and algorithm design, large-scale automated explorations of chemical reaction space have become possible. Here, we present our approach to an open-source, extensible framework for explorations of chemical reaction mechanisms based on the first-principles of quantum mechanics. It is intended to facilitate reaction network explorations for diverse chemical problems with a wide range of goals such as mechanism elucidation, reaction path optimization, retrosynthetic path validation, reagent design, and microkinetic modeling. The stringent first-principles basis of all algorithms in our framework is key for the general applicability that avoids any restrictions to specific chemical systems. Such an agile framework requires multiple specialized software components of which we present three modules in this work. The key module, Chemoton, drives the exploration of reaction networks. For the exploration itself, we introduce two new algorithms for elementary-step searches that are based on Newton trajectories. The performance of these algorithms is assessed for a variety of reactions characterized by a broad chemical diversity in terms of bonding patterns and chemical elements. Chemoton successfully recovers the vast majority of these. We provide the resulting data, including large numbers of reactions that were not included in our reference set, to be used as a starting point for further explorations and for future reference.
BibTeX
@article{unsleber2022chemoton,
  author = {Jan P. Unsleber and Stephanie A. Grimmel and Markus Reiher},
  title = {{Chemoton 2.0: Autonomous Exploration of Chemical Reaction Networks}},
  journal = {Journal of Chemical Theory and Computation},
  year = {2022},
  doi = {10.1021/acs.jctc.2c00193},
}
Figure