ACS Central Science最新文献

Metal–organic frameworks (MOFs) are a major targetof machine-learning-basedproperty prediction, yet most models assume that a single frameworkrepresentation maps to a single property value. This assumption becomesproblematic for experimental MOFs, where samples reported as the sameframework can exhibit different properties because of differencesin crystallinity, phase purity, defects, and other sample-dependentfactors. Here we introduce Experimental X-ray Diffraction IntegratedTransformer (EXIT), a multimodal transformer for sample-aware predictionof MOF properties that combines MOFid with X-ray diffraction (XRD).In EXIT, MOFid encodes MOF identity, whereas XRD provides complementaryinformation about the experimentally realized sample state. EXIT ispretrained on one million hypothetical MOFs with simulated XRD tolearn transferable representations. The pretrained model is firstevaluated through fine-tuning on downstream tasks using simulatedXRD, where it achieves improved performance relative to existing approaches,and is subsequently fine-tuned on literature-derived experimentalXRD–property data sets for surface area and pore volume prediction.Incorporating experimental XRD improves predictive performance relativeto models without experimental XRD, and attention analysis and sample-levelcase studies further show that EXIT assigns different predictionsto samples sharing the same MOF identity when their XRD patterns differ.These results establish a practical step from framework-aware to sample-awareMOF property prediction and highlight the value of incorporating experimentalcharacterization into porous materials informatics.

Complex organic molecules in interstellar ices and carbonaceousasteroids record key steps in the chemical evolution toward life,yet the origin of branched carbonyl compounds─a fundamentalclass of biorelevant molecules─remains unresolved. Here, wedemonstrate the efficient formation of C₄ carbonyls, isobutyraldehyde((CH₃)₂CHCHO) and 2-butanone (CH₃CH₂COCH₃) via barrierless radical–radicalrecombination in carbon monoxide–propane (CO–C₃H₈; 1:1.1 ± 0.2) and ethane–acetaldehyde (C₂H₆–CH3CHO; 1.5 ± 0.3:1) ice mixturesat 5 K irradiated with 5 keV electrons as proxies for secondary electronsgenerated by galactic cosmic rays. Isobutyraldehyde arises from recombinationof formyl (HĊO) and isopropyl (CH₃ĊHCH₃) radicals, whereas 2-butanone forms through acetyl (CH₃ĊO) and ethyl (CH₃ĊH₂)radical coupling. Using isomer-selective photoionization mass spectrometrywith isotopic labeling, we provided strong evidence for these productstogether with the enol 2-methylprop-1-en-1-ol ((CH₃)₂CCHOH) in the gas phase. These results establish a plausiblelow-temperature mechanism for molecular mass growth that generatesbranched carbon skeletons without activation barriers, bridging acritical gap between simple interstellar species and structurallycomplex, biorelevant organics. The demonstrated pathways operate undercosmic-ray-driven, nonequilibrium chemistry in icy grains, providinga plausible route to C₄ backbone motifs found in prebioticmolecules, including amino acids and fatty acids. Our findings showthat chemical complexity─including carbon skeleton branching─canemerge in deep space prior to planetary accretion, implying a plausibleextraterrestrial origin for key molecular precursors delivered toearly Earth and exo planetary systems.

Large language models(LLMs) are becoming increasingly appliedbeyond natural language processing, demonstrating strong capabilitiesin complex scientific tasks that traditionally require human expertise.This progress has extended into materials discovery, where LLMs introducea new paradigm by leveraging reasoning and in-context learning, capabilitiesabsent from conventional machine learning approaches. Here, we presenta Multi-Agent-based Electrocatalyst Search Through Reasoning and Optimization(MAESTRO) framework in which multiple LLMs with specialized rolescollaboratively discover high-performance single-atom catalysts forthe oxygen reduction reaction. Within an autonomous design loop, agentsiteratively reason, propose modifications, reflect on results, andaccumulate design history. Through in-context learning enabled bythis iterative process, MAESTRO identified design principles not explicitlyencoded in the LLMs’ background knowledge and successfullydiscovered catalysts that break conventional scaling relations betweenreaction intermediates. These results highlight the potential of multiagentLLM frameworks as a powerful strategy to generate chemical insightand discover promising catalysts.

Spectroscopy basedon individual-photon-level light sources bringsnew variables to both spectroscopy and imaging. We refer to this approachas quantum-light spectroscopy (QLS). The photon statistics of differenttypes of photon sources and photon entanglements are briefly described.We then illustrate the potential of QLS by means of some current experimentalexamples including time-resolved spectroscopy performed with singlephotons and quantum-enhanced optical imaging, showing how all themethods rely on correlations between entangled photons in the pair.We then ask: how might physical chemists exploit QLSs to expose newphysical insights? For example, can we address fundamental questionsin interactions between light and molecules or materials that arevery difficult to address with classical light sources? Does QLS openup techniques to study highly congested spectra with higher combinedspectral and temporal resolution than is possible with classical light?The concept of QLS also opens the potential study of fundamental questionssuch as what initial excited states are produced by individual photonabsorption? We also point out the experimental challenges in extendingQLS to more complex spectroscopies, such as multidimensional spectroscopies.We conclude with a brief selection of recent theoretical proposalsto extend the capabilities of the QLS.

Organic cations andradicals are promising building componentsfor organic optoelectronics, spintronics and quantum information processing.Realizing their full potential requires strategies to enhance theirstability and precisely control their solid-state organization. Herein,we describe a modular synthetic strategy via photoelectrocyclizationin the presence of tetracyanoethylene (TCNE) to access a series ofclosed-shell phenalenyl-derived cations featuring an oxonium-embeddedbenzo[b]perylene (OBP⁺) π-skeleton. The resulting OBP⁺ cations can be reversibly reduced to generate neutralopen-shell phenalenyl-derived radicals. Noteworthily, we discovereda single crystal of OBP3·TCNE^• that readily formed from the reaction mixture, displayingone-dimensional (1D) π-stacked arrays of alternating closed-shell OBP3⁺ cation and open-shell TCNE^•– radical anion. This solid-state arrangementstabilizes otherwise reactive TCNE^•– species,preserving its open-shell character even after one month of ambientexposure. OBP3·TCNE^• is a rare example of purely organic paramagnetic salt and has beensystematically investigated using integrated spectroscopic, magnetic,and computational analyses, together with preliminary photoconductivitymeasurements (ϕΣμ = 1.4 × 10^–5 cm² V^–1 s^–1). Overall, this work heralds a versatile photochemicalroute to redox-switchable cationic π-scaffolds and persistentopen-shell crystalline organic materials.

Low-density lipoproteins (LDL) are naturally occurring,monodispersenanoparticles composed of proteins and lipids with an average diameterof 18–25 nm. Their biological functions such as cholesteroltrafficking and their role in various diseases have been extensivelystudied. However, their properties as nanoparticles are seldom discussed.Here, we show that when isolated from biofluids, analogous to anysynthetic nanoparticle, LDL carry a distinctive protein corona. Thecorona can be partly removed or preserved depending on the approachused to isolate them. Importantly, we find that proteomic analysisof this corona enables proof-of-concept separation between healthyindividuals and patients with melanoma, colon carcinoma, or livercancer. This approach provided clearer separation than direct proteomicanalysis of serum because the corona of the LDL particle is relativelydepleted of the serum-abundant proteins, most of which are only minorcomponents of the corona. We postulate that proteomic analysis oflipoproteins can reveal disease-associated changes and that a deepunderstanding of the corona of these particles will advance the overallfield of nanoparticles in medicine.