Accurate prediction of the physicochemical properties of molecular mixtures using graph neural networks remains a significant challenge, as it requires simultaneous embedding of intramolecular interactions while accounting for mixture composition (i.e., concentrations and ratios). Existing approaches are ill-equipped to emulate realistic mixture environments, where densely coupled interactions propagate across hierarchical levels - from atoms and functional groups to entire molecules - and where cross-level information exchange is continuously modulated by composition. To bridge the gap between isolated molecules and realistic chemical environments, we present ChemFlow, a novel hierarchical framework that integrates atomic, functional group, and molecular-level features, facilitating information flow across these levels to predict the behavior of complex chemical mixtures. ChemFlow employs an atomic-level feature fusion module, Chem-embed, to generate context-aware atomic representations influenced by the mixture state and atomic characteristics. Next, bidirectional group-to-molecule and molecule-to-group attention mechanisms enable ChemFlow to capture functional group interactions both within and across molecules in the mixture. By dynamically adjusting representations based on concentration and composition, ChemFlow excels at predicting concentration-dependent properties and significantly outperforms state-of-the-art models in both concentration-sensitive and concentration-independent systems. Extensive experiments demonstrate ChemFlow's superior accuracy and efficiency in modeling complex chemical mixtures.

Semiartificial photosynthesis exploits synergies between synthetic light absorbers and biological catalysts, offering a promising strategy for solar chemistry. Unlike conventional synthetic catalysts for carbon dioxide (CO2) fixation, biological systems employ enzymes, most notably formate dehydrogenases (Fdhs), to catalyze the interconversion between CO2, protons and electrons into formate, a central hub molecule in energy and carbon metabolism. This review focuses on the deployment of metal-dependent Fdhs in semiartificial photosynthesis, with an emphasis on molybdenum- and tungsten-dependent enzymes directly wired to electrodes and synthetic light absorbers. We first examine the structural, mechanistic, and redox properties of relevant Fdhs in vivo and in vitro, highlighting reaction pathways and inherent challenges. Subsequent sections discuss the central role of biotic-abiotic interfaces in constructing functional biohybrid systems, highlighting how advanced interfacial characterization techniques inform enzyme loading, charge carrier dynamics, and reaction intermediates. We then summarize progress and challenges in (photo)electrochemical and photochemical systems leveraging Fdhs unique properties as a model catalyst for CO2-to-formate conversion. This review aims to clarify the current state of the semiartificial photosynthesis field employing metal-dependent Fdhs in vitro, and guide future research at the interface of enzymology, photo(electro)chemistry, and materials science.

Despite great industrial interest, modeling the physical properties of surfactants in water based on their molecular structure remains a challenge. A significant part of this challenge is in obtaining sufficient amounts of high-quality data. Experimentally determined properties such the critical micelle concentration (CMC) and surface tension at CMC (gamma CMC) have been reported for many surfactants. However, surfactant data are scattered across many literature sources, and reported in a manner which is often unsuitable as input for predictive models. In this work, we address this limitation by compiling the SurfPro database of surfactant properties. SurfPro consists of 1624 surfactant entries curated from 223 literature sources, containing 1395 CMC values, 972 gamma CMC values and more than 657 values for Gamma max, C20, pi CMC and Amin. However, only 647 structures have all reported properties, and for most surfactants multiple properties are missing. We trained a previously reported graph neural network architecture for single- and multi-property prediction on these incomplete data of all surfactant types in the database to accurately predict pCMC (-log10(CMC)), gamma CMC, Gamma max and pC20. We achieved state-of-the-art performance of these four properties using an ensemble of AttentiveFP models trained on ten different folds of the training data in the multi-property setting. Finally, we leveraged the predictions and uncertainties of the ensemble model to impute all missing properties for all 977 surfactants with an incomplete set of properties. We make our curated SurfPro database, proposed test split and training datasets, the imputed database, as well as our code publicly available.

It is generally accepted that minerals were an important source of prebiotic catalysis. In this work we demonstrate how the prebiotic sugar forming formose reaction is guided to unique reaction compositions in the presence of a variety of minerals. When the same mineral is transferred between multiple sequential batch reactions, a new reaction composition is obtained after each reaction cycle. We attribute this effect to the adsorption of catalytic Ca(OH)2 to mineral surfaces. Further exploration shows that first exposing the mineral surface to the aqueous catalyst allows the mineral to subsequently produce formose outputs without the need for any additional catalyst to be present. As such, the mineral surface functions as storage of the preceding environmental conditions. Our work supports the development of chemical complexity through the transfer of information between sequences of chemical environments.

In chemical reaction network theory, ordinary differential equations are used to model the temporal change of chemical species concentration. As the functional form of these ordinary differential equation systems is derived from an empirical model of the reaction network, it may be incomplete. Our approach aims to elucidate these hidden insights in the reaction network by combining dynamic modeling with deep learning in the form of neural ordinary differential equations. Our contributions not only help to identify the shortcomings of existing empirical models but also assist the design of future reaction networks.

Chemical language models can be fine-tuned to predict properties, like water solubility, given molecular inputs. This work investigates how these predictions may be explained.

Chemical reaction networks, such as those found in metabolism and signalling pathways, enable cells to process information from their environment1,2. Current approaches to molecular information processing and computation typically pursue digital computation models and require extensive molecular-level engineering3. Despite considerable advances, these approaches have not reached the level of information processing capabilities seen in living systems. Here we report on the discovery and implementation of a chemical reservoir computer based on the formose reaction4. We demonstrate how this complex, self-organizing chemical reaction network can perform several nonlinear classification tasks in parallel, predict the dynamics of other complex systems and achieve time-series forecasting. This in chemico information processing system provides proof of principle for the emergent computational capabilities of complex chemical reaction networks, paving the way for a new class of biomimetic information processing systems.

Explainability techniques are crucial in gaining insights into the reasons behind the predictions of deep learning models, which have not yet been applied to chemical language models. We propose an explainable AI technique that attributes the importance of individual atoms towards the predictions made by these models. Our method backpropagates the relevance information towards the chemical input string and visualizes the importance of individual atoms. We focus on self-attention Transformers operating on molecular string representations and leverage a pretrained encoder for finetuning. We showcase the method by predicting and visualizing solubility in water and organic solvents. We achieve competitive model performance while obtaining interpretable predictions, which we use to inspect the pretrained model.

Prebiotic environments are dynamic, containing a range of periodic and aperiodic variations in reaction conditions. However, the impact of the temporal dynamics of environmental conditions upon prebiotic chemical reaction networks has not been investigated. Here, we demonstrate how the magnitude and rate of temporal fluctuations of the catalysts Ca2+ and hydroxide control the product distributions of the formose reaction. Surprisingly, the product compositions of the formose reaction under dynamic conditions deviate significantly from those under steady state conditions. We attribute these compositional changes to the non-uniform propagation of fluctuations through the network, thereby shaping reaction outcomes. An examination of temporal concentration patterns showed that collections of compounds responded collectively to perturbations, indicating that key gating reactions branching from the Breslow cycle may be important responsive features of the formose reaction. Our findings show how the compositions of prebiotic reaction networks were shaped by sequential environmental events, illustrating the necessity for considering the temporal traits of prebiotic environments that supported the origin of life.

The evolution of life from the prebiotic environment required a gradual process of chemical evolution towards greater molecular complexity. Elaborate prebiotically relevant synthetic routes to the building blocks of life have been established. However, it is still unclear how functional chemical systems evolved with direction using only the interaction between inherent molecular chemical reactivity and the abiotic environment. Here we demonstrate how complex systems of chemical reactions exhibit well-defined self-organization in response to varying environmental conditions. This self-organization allows the compositional complexity of the reaction products to be controlled as a function of factors such as feedstock and catalyst availability. We observe how Breslow's cycle contributes to the reaction composition by feeding C2 building blocks into the network, alongside reaction pathways dominated by formaldehyde-driven chain growth. The emergence of organized systems of chemical reactions in response to changes in the environment offers a potential mechanism for a chemical evolution process that bridges the gap between prebiotic chemical building blocks and the origin of life.

Abstract Prebiotic environments were dynamic, containing a range of periodic and aperiodic variations in reaction conditions. However, the impact of the temporal dynamics of environmental conditions upon prebiotic chemical reaction networks (CRNs) has not been investigated. Here, we demonstrate how the magnitude and rate of temporal fluctuations of the catalysts Ca2+ and OH– control product distributions of the formose reaction. Surprisingly, the product compositions of the formose reaction under dynamic conditions deviate significantly from those under steady state conditions. We attribute these compositional changes to the non-uniform propagation of fluctuations through the network thereby shaping reaction outcomes. Examination of temporal concentration patterns showed that collections of compounds responded collectively to perturbations, indicating key ‘gating’ reactions branching from the Breslow cycle may be important responsive features of the formose reaction. Our findings show how the compositions of prebiotic reaction networks were shaped by sequential environmental events, illustrating the necessity for considering the temporal traits of prebiotic environments that supported the origin of life.

Metal-dependent formate dehydrogenases (FDHs) catalyze the reversible conversion of formate into CO2, a proton, and two electrons. Kinetic studies of FDHs provide key insights into their mechanism of catalysis, relevant as a guide for the development of efficient electrocatalysts for formate oxidation as well as for CO2 capture and utilization. Here, we identify and explain the kinetic isotope effect (KIE) observed for the oxidation of formate and deuterioformate by the Mo-containing FDH from Escherichia coli using three different techniques: steady-state solution kinetic assays, protein film electrochemistry (PFE), and pre-steady-state stopped-flow methods. For each technique, the Mo center of FDH is reoxidized at a different rate following formate oxidation, significantly affecting the observed kinetic behavior and providing three different viewpoints on the KIE. Steady-state turnover in solution, using an artificial electron acceptor, is kinetically limited by diffusional intermolecular electron transfer, masking the KIE. In contrast, interfacial electron transfer in PFE is fast, lifting the electron-transfer rate limitation and manifesting a KIE of 2.44. Pre-steady-state analyses using stopped-flow spectroscopy revealed a KIE of 3 that can be assigned to the C-H bond cleavage step during formate oxidation. We formalize our understanding of FDH catalysis by fitting all the data to a single kinetic model, recreating the condition-dependent shift in rate-limitation of FDH catalysis between active-site chemical catalysis and regenerative electron transfer. Furthermore, our model predicts the steady-state and time-dependent concentrations of catalytic intermediates, providing a valuable framework for the design of future mechanistic experiments.

AbstractThe integration of enzymes with synthetic materials allows efficient electrocatalysis and production of solar fuels. Here, we couple formate dehydrogenase (FDH) from Desulfovibrio vulgaris Hildenborough (DvH) to metal oxides for catalytic CO2 reduction and report an in‐depth study of the resulting enzyme–material interface. Protein film voltammetry (PFV) demonstrates the stable binding of FDH on metal‐oxide electrodes and reveals the reversible and selective reduction of CO2 to formate. Quartz crystal microbalance (QCM) and attenuated total reflection infrared (ATR‐IR) spectroscopy confirm a high binding affinity for FDH to the TiO2 surface. Adsorption of FDH on dye‐sensitized TiO2 allows for visible‐light‐driven CO2 reduction to formate in the absence of a soluble redox mediator with a turnover frequency (TOF) of 11±1 s−1. The strong coupling of the enzyme to the semiconductor gives rise to a new benchmark in the selective photoreduction of aqueous CO2 to formate.

ConspectusEnzymes are the essential catalytic components of biology and adsorbing redox-active enzymes on electrode surfaces enables the direct probing of their function. Through standard electrochemical measurements, catalytic activity, reversibility and stability, potentials of redox-active cofactors, and interfacial electron transfer rates can be readily measured. Mechanistic investigations on the high electrocatalytic rates and selectivity of enzymes may yield inspiration for the design of synthetic molecular and heterogeneous electrocatalysts. Electrochemical investigations of enzymes also aid in our understanding of their activity within their biological environment and why they evolved in their present structure and function. However, the conventional array of electrochemical techniques (e.g., voltammetry and chronoamperometry) alone offers a limited picture of the enzyme–electrode interface.How many enzymes are loaded onto an electrode? In which orientation(s) are they bound? What fraction is active, and are single or multilayers formed? Does this static picture change over time, applied voltage, or chemical environment? How does charge transfer through various intraprotein cofactors contribute to the overall performance and catalytic bias? What is the distribution of individual enzyme activities within an ensemble of active protein films? These are central questions for the understanding of the enzyme–electrode interface, and a multidisciplinary approach is required to deliver insightful answers.Complementing standard electrochemical experiments with an orthogonal set of techniques has recently allowed to provide a more complete picture of enzyme–electrode systems. Within this framework, we first discuss a brief history of achievements and challenges in enzyme electrochemistry. We subsequently describe how the aforementioned challenges can be overcome by applying advanced electrochemical techniques, quartz-crystal microbalance measurements, and spectroscopic, namely, resonance Raman and infrared, analysis. For example, rotating ring disk electrochemistry permits the simultaneous determination of reaction kinetics and quantification of generated products. In addition, recording changes in frequency and dissipation in a quartz crystal microbalance allows to shed light into enzyme loading, relative orientation, clustering, and denaturation at the electrode surface. Resonance Raman spectroscopy yields information on ligation and redox state of enzyme cofactors, whereas infrared spectroscopy provides insights into active site states and the protein secondary and tertiary structure. The development of these emerging methods for the analysis of the enzyme–electrode interface is the primary focus of this Account. We also take a critical look at the remaining gaps in our understanding and challenges lying ahead toward attaining a complete mechanistic picture of the enzyme–electrode interface.

The biological formate hydrogenlyase (FHL) complex links a formate dehydrogenase (FDH) to a hydrogenase (H2ase) and produces H2 and CO2 from formate via mixed-acid fermentation in Escherichia coli. Here, we describe an electrochemical and a colloidal semiartificial FHL system that consists of an FDH and a H2ase immobilized on conductive indium tin oxide (ITO) as an electron relay. These in vitro systems benefit from the efficient wiring of a highly active enzyme pair and allow for the reversible conversion of formate to H2 and CO2 under ambient temperature and pressure. The hybrid systems provide a template for the design of synthetic catalysts and surpass the FHL complex in vivo by storing and releasing H2 on demand by interconverting CO2/H2 and formate with minimal bias in either direction.

Solar-driven coupling of water oxidation with CO2 reduction sustains life on our planet and is of high priority in contemporary energy research. Here, we report a photoelectrochemical tandem device that performs photocatalytic reduction of CO2 to formate. We employ a semi-artificial design, which wires a W-dependent formate dehydrogenase (FDH) cathode to a photoanode containing the photosynthetic water oxidation enzyme, Photosystem II, via a synthetic dye with complementary light absorption. From a biological perspective, the system achieves a metabolically inaccessible pathway of light-driven CO2 fixation to formate. From a synthetic point of view, it represents a proof-of-principle system utilizing precious-metal-free catalysts for selective CO2-to-formate conversion using water as an electron donor. This hybrid platform demonstrates the translatability and versatility of coupling abiotic and biotic components to create challenging models for solar fuel and chemical synthesis.

Natural photosynthesis stores sunlight in chemical energy carriers, but it has not evolved for the efficient synthesis of fuels, such as H2. Semi-artificial photosynthesis combines the strengths of natural photosynthesis with synthetic chemistry and materials science to develop model systems that overcome nature’s limitations, such as low-yielding metabolic pathways and non-complementary light absorption by photosystems I and II. Here, we report a bias-free semi-artificial tandem platform that wires photosystem II to hydrogenase for overall water splitting. This photoelectrochemical cell integrated the red and blue light-absorber photosystem II with a green light-absorbing diketopyrrolopyrrole dye-sensitized TiO2 photoanode, and so enabled complementary panchromatic solar light absorption. Effective electronic communication at the enzyme–material interface was engineered using an osmium-complex-modified redox polymer on a hierarchically structured TiO2. This system provides a design protocol for bias-free semi-artificial Z schemes in vitro and provides an extended toolbox of biotic and abiotic components to re-engineer photosynthetic pathways. Semi-artificial photosynthetic systems combine natural and synthetic features to overcome limitations of each approach to produce solar fuels. Sokol et al. integrate a dye-sensitized TiO2 photoanode with the natural machineries, photosystem II and hydrogenase, to split water without additional applied bias.

Molybdenum-containing formate dehydrogenase H from Escherichia coli (EcFDH-H) is a powerful model system for studies of the reversible reduction of CO2 to formate. However, the mechanism of FDH catalysis is currently under debate, and whether the primary Mo coordination sphere remains saturated or one of the ligands dissociates to allow direct substrate binding during turnover is disputed. Herein, we describe how oxidation-state-dependent changes at the active site alter its inhibitor binding properties. Using protein film electrochemistry, we show that formate oxidation by EcFDH-H is inhibited strongly and competitively by N3-, OCN-, SCN-, NO2-, and NO3-, whereas CO2 reduction is inhibited only weakly and not competitively. During catalysis, the Mo center cycles between the formal Mo(VI)═S and Mo(IV)-SH states, and by modeling chronoamperometry data recorded at different potentials and substrate and inhibitor concentrations, we demonstrate that both formate oxidation and CO2 reduction are inhibited by selective inhibitor binding to the Mo(VI)═S state. The strong dependence of inhibitor-binding affinity on both Mo oxidation state and inhibitor electron-donor strength indicates that inhibitors (and substrates) bind directly to the Mo center. We propose that inhibitors bind to the Mo following dissociation of a selenocysteine ligand to create a vacant coordination site for catalysis and close by considering the implications of our data for the mechanisms of formate oxidation and CO2 reduction.