Ganna Gryn'OVa

Topological nets define the connectivity in framework materials, influencing their pore structure and mechanical properties. In this work, we investigate the importance of incorporating this information into machine learning models for gas adsorption prediction. To this end, we develop a novel representation, persistent homology using nets (PHuN), which computes topological features directly from the framework net via persistent homology. We show that PHuN leads to a computationally efficient representation that captures important topological information. We apply this representation to predict gas adsorption properties and demonstrate its ability to produce accurate predictions, especially when combined with other commonly used geometric and chemical descriptors. Further, to gain interpretable insights, we perform an ablation study combined with an analysis of feature importance to evaluate the contribution of different sets of descriptors and inspect how topological, geometrical, and chemical descriptors affect model performance.

Here we present a solvent decision-making pathway for less hazardous and scalable synthesis of TpPa-1 covalent organic framework (COF) underpinned by environmental compatibility (CHEM21 framework), processability (Hansen solubility parameters), and thermodynamic suitability (free energy of solvation). Our analyses suggest that solvents with moderate solvation, such as 1,4-dioxane, propylene carbonate, and diacetin, hinder excessive reaction reversibility and promote crystallite growth. Balancing solvent greenness, compatibility, and thermodynamics, we find that diacetin is optimal for a continuous-flow synthesis. Compared with TpPa-1 COF synthesized in diacetin under batch conditions, the Brunauer–Emmett–Teller surface area of flow-derived TpPa-1 reached 418 m² g^–1, increasing CO₂ uptake by 50% at 298 K. The flow process achieved a 30-fold increase in space-time yield (STY), with excellent reproducibility and an 89% reduction in specific energy consumption. This solvent-thermodynamics-guided approach establishes a physically transparent and sustainable framework for optimizing COF synthesis under continuous-flow conditions.

Reduction potentials of redox-active molecules and materials are essential descriptors of their performance as catalysts, antioxidants, electrode materials, etc. For a given species, its practical applications often span a range of solvent environments, which profoundly impact its redox properties. In this work, we present a message passing graph neural network architecture with a Set Transformer readout trained on ca. 20,000 reduction potentials of chemically diverse closed- and open-shell organic redox-active molecules (the "ReSolved" data set), computed using a rigorously benchmarked density functional theory procedure. The predictor model affords high accuracy with mean absolute errors of ca. 0.2 eV and is uniquely able to generalize to previously unseen solvents. We couple this architecture with an evolutionary algorithm to inverse-design synthetically accessible candidate molecules with target reduction potentials for several battery-related practical applications.

Using a new automatic fragmentation algorithm, deCOFpose , we elucidate how topologies and building blocks determine the band gaps of covalent organic frameworks.

Reduction potentials of redox-active molecules and materials are essential descriptors of their performance as catalysts, antioxidants, electrode materials, etc. For a given species, its practical applications often span a range of solvent environments, which profoundly impact its redox properties. In this work, we present a message passing graph neural network architecture with a set transformer readout trained on ca. 20,000 reduction potentials of chemically diverse closed- and open-shell organic redox-active molecules (the “ReSolved” dataset), computed using a rigorously benchmarked density functional theory procedure. The predictor model affords high accuracy with mean absolute errors of ca. 0.2 eV and is uniquely able to generalise to previously unseen solvents. We couple this architecture with an evolutionary algorithm to inverse-design synthetically accessible candidate molecules with target reduction potentials for several battery-related practical applications.

Understanding structure-property relationships in ordered functional materials is essential for their rational design and optimisation. Fragment-based approaches relating materials’ properties to those of their building blocks (fragments) are intuitive to chemistry and have been successfully applied in the design of metal-organic frameworks (MOFs). However, covalent organic frameworks (COFs) are resistant to such in silico fragmentation due to their covalent bonds and ambiguous definitions of nodes and linkers. Here we introduce a new algorithm, deCOFpose, designed to systematically fragment COFs into building blocks according to chemically intuitive rules, enabling fragment-based structure-property analysis, and exemplify the latter for COFs band gaps. Our results reveal that the electronic features (e.g., energies of the frontier molecular orbitals) of the building blocks alone are insufficient to fully represent these materials, and the inclusion of their topological characteristics is required to engineer bespoke COFs with desired band structures.

The photodegradation of ketones, including polyketones, by Norrish degradation pathways is well established. However, these mechanistic pathways lead to uncontrolled reactions, which in the case of plastics, result in in-complete degradation and the formation of microplastic fragments. Using an organocatalyzed thiol-yne step-growth polymerization, we developed poly(enone sulfides) that display useful properties and undergo a highly selective photodegradation into non-cytotoxic diketone and disulfide products. Model substrate reactions, sup-ported by computational modelling, revealed the molecular features that enable the selective degradative path-way to occur, implicating both radical and reactive oxygen species. The synthetic approaches for monomers and polymers are versatile and allow for the design of a range of polymer structures with divergent thermomechanical properties, with some photodegradable materials being tough thermoplastics possessing high strength and ductil-ity.

Hydrogen atom transfer (HAT) reactions are important in many biological systems. As these reactions are hard to observe experimentally, it is of high interest to shed light on them using simulations. Here, we present a machine learning model based on graph neural networks for the prediction of energy barriers of HAT reactions in proteins. As input, the model uses exclusively non-optimized structures as obtained from classical simulations. It was trained on more than 17 000 energy barriers calculated using hybrid density functional theory. We built and evaluated the model in the context of HAT in collagen, but we show that the same workflow can easily be applied to HAT reactions in other biological or synthetic polymers. We obtain for relevant reactions (small reaction distances) a model with good predictive power (R² ∼ 0.9 and mean absolute error of −1). As the inference speed is high, this model enables evaluations of dozens of chemical situations within seconds. When combined with molecular dynamics in a kinetic Monte-Carlo scheme, the model paves the way toward reactive simulations.

Hydrogen atom transfer (HAT) reactions are important in many biological systems. As these reactions are hard to observe experimentally, it is of high interest to shed light on them using simulations. Here, we present a machine learning model based on graph neural networks for the prediction of energy barriers of HAT reactions in proteins. As input, the model uses exclusively non-optimized structures as obtained from classical simulations. It was trained on more than 17,000 energy barriers calculated using hybrid density functional theory. We built and evaluated the model in the context of HAT in collagen, but we show that the same workflow can easily be applied to HAT reactions in other biological or synthetic polymers. We obtain for relevant reactions (small reaction distances) a model with good predictive power (R2 ∼ 0.9 and mean absolute error of <3 kcal/mol). As the inference speed is high, this model enables evaluations of dozens of chemical situations within seconds. When combined with molecular dynamics in a kinetic Monte-Carlo scheme, the model paves the way toward reactive simulations.