Articles in 2023

Gas chromatography is a useful tool to identify and characterize wines, usually by selecting some compounds for a particular classification problem, yet, with limited success. Here, the authors decode the estates perfectly and age 50% correctly of twelve red Bordeaux wines from unrestricted, raw gas chromatograms using machine learning.

Artificial photosynthesis aims to produce fuels and chemicals from simple, abundant building blocks, such as water and carbon dioxide, with sunlight as a source of energy. Here, the authors review recent developments in biomimetic, compartmentalized vesicular systems towards artificial photosynthesis, and highlight challenges and opportunities in mimicking this complex natural reaction system.

Non-cell-permeable cryoprotectants exhibit cryopreserving effects by preventing intracellular ice crystal formation, but these can damage cells due to high osmotic pressure and dehydration. Here, the authors developed a poly(zwitterion) isotonic cryoprotectant that forms a firm matrix around cells that prevents the influx of ice crystal nuclei without the need for high osmotic pressure and dehydration.

Atom-centered neural networks represent the state-of-the-art for approximating quantum chemical properties of molecules, such as internal energies, but the final atom pooling operation that is necessary to convert from atomic to molecular representations in most models remains relatively undeveloped. Here, the authors report a learnable pooling operation, usable as a drop-in replacement, that leverages an attention mechanism to model interactions between atom representations.

The variable frequency microwave (VFM) technology is known for its theoretically uniform microwave E-field distribution, although the experimental distribution remains poorly understood. Here, the authors examine the effectiveness of VFM irradiation experimentally by in situ 3D observation of the E-field, and apply VFM to the large scale synthesis of 4-methylbiphenyl via the Suzuki-Miyaura cross-coupling reaction catalyzed by solid Pd supported on activated carbon particulates.

Shortwave UV photons and very low energy electrons (vLEEs) are thought to be unfavorable prebiotic conditions on early Earth which can destroy unstable molecules. Here, the authors propose that nucleobases in their complementary pairs can enhance and consolidate the intrinsic stability of nucleobases against shortwave UV photons and vLEEs and promote their proliferation.

Seleno-insulins (SeIns) are known to exhibit different in vitro properties from wild-type insulin, upon replacement of one of the three disulfide bonds with diselenide bonds; however, the in vivo hypoglycemic effect remains poorly understood. Here, the authors optimize the synthesis of SeIns via native chain assembly and reveal in vitro resistance against the insulin-degrading enzyme, as well as the in vivo hypoglycemic effect in diabetic rats at a dose of 150 μg/300 g rat.

Late-stage functionalization of drug molecules can tune their properties without the need for entirely new syntheses, however, predicting reactivity and planning synthesis for late-stage C-H activation remains challenging. Here, the authors develop a reaction screening approach combining high-throughput experimentation with computational graph neural networks to identify suitable substrates that can be used for late-stage C-H alkylation via Minisci-type chemistry.

β-Amino acid-containing macrolactams are a known natural product family exhibiting structural and functional diversity, however, the natural chemical space of this family remains underexplored. Here, the authors use a targeted β-amino acid-specific homology-based multi-query search to identify their potential microbial producers, explore the variation of their biosynthetic gene clusters, heterologously produce ciromicin A, and identify new macrotermycin derivatives.

The reduction of dinitrogen to ammonia catalyzed by nitrogenase involves long-range conformational changes within the enzyme complex, however, direct biophysical evidence of communication between the Fe protein and the MoFe protein is lacking. Here, the authors combine millisecond time-resolved hydrogen-deuterium exchange mass spectrometry and normal mode analysis, revealing molecular-level insights into how the Fe protein alters the stability and dynamics of the MoFe protein near the active site in a nucleotide-dependent manner.

Tetraphenylethylenes (TPEs) are known for their aggregation-induced emission and electrochemical properties of value for optical sensors, however, established synthetic routes suffer from several limitations. Here, the authors develop a metal-free route to TPEs and other alkenes via a sulfur-mediated coupling of N-tosylhydrazones, achieving good yields and a broad substrate scope.

Mineral precipitation in porous media is relevant for various energy-related applications such as oil extraction, geothermal systems, or nuclear waste disposal, but crystallization retardation as a function of pore size and shape is poorly explored. Here, the authors study barite crystallization in micro-confinement, showing that retardation of crystallization can start in pores of less than 1 µm in size, with the probability of nucleation scaling with pore volume.

Chemical variational autoencoder (VAE) is a deep learning method for constructing chemical latent spaces, however, the current chemical VAE variants are limited in their ability to handle complex compound structures like natural products. Here, the authors develop a variational autoencoder that can handle natural products and demonstrate its utility as an in silico drug discovery tool.

The bio-hybrid fuel 4,5-dimethyl-1,3-dioxolane can be produced from glucose via 2,3-butanediol (2,3-BDO), however, the separation of microbially produced 2,3-BDO remains costly. Here, the authors develop an alternative process route to 4,5-dimethyl-1,3-dioxolane via the enzymatic production of 2,3-BDO.

Gadolinium-based contrast agents are the gold standard for magnetic resonance imaging (MRI), however, their poor stability causes safety problems in clinical applications. Here, the authors develop two chiral Gd(III) DOTA-based complexes with a macrocyclic backbone that show higher relaxivities and stabilities than benchmark complexes.

Cu(I)-thioether complexes have been widely studied as metalloprotein mimics, but limited structural geometries have been reported to date. Here, the authors demonstrate that a chiral cyclic β-amino acid ligand affords Cu(I)-thioether complexes with chiral conformations and the capacity for hydrogen bonding, offering unusual geometric structures and metal-to-ligand ratios.

Achieving kinetic control over reversible self-assembly processes remains challenging. Here, the authors assemble Rh₄L₄ coordination squares consisting of cis-protected Rh(II) corner sites and linear ditopic ligands under kinetic control, preventing the formation of triangles by modulating the energy landscape with use of a leaving ligand.

Supramolecular networks deposited on surfaces are of interest for a range of applications, but expanding monolayers to well-ordered multilayers remains a significant challenge, not least because analytical tools capable of probing such assemblies are limited. Here, the authors demonstrate that combining AFM-IR with STM enables an assessment of the differences in molecular conformations between 2D and 3D supramolecular networks adsorbed onto a HOPG surface.

Benzil and its derivatives have been widely studied for their luminescence properties, but the effects of ortho- and meta-substituents on photophysical properties have yet to be explored. Here, ortho-, meta- and para-substituted benzil derivatives are accessed via an organocatalytic pinacol coupling of substituted benzaldehydes, and ortho- and meta-substituents are shown to significantly influence the electronic state and luminescence properties of benzil.

The endeavor to harness machine learning to improve antibody-antigen interaction is not only significant but also fraught with challenges due to the limitations of the publicly available datasets. Here, the authors introduce an Antibody Random Forest Classifier that classifies mutations as either deleterious or non-deleterious based on physicochemical features, which enabled the identification of affinity-enhancing mutations in SARS-CoV-2 antibodies with an up to 1000-fold increased binding affinity.