Prediction rigidities for data-driven chemistry.

The widespread application of machine learning (ML) to the chemical sciences is making it very important to understand how the ML models learn to correlate chemical structures with their properties, and what can be done to improve the training efficiency whilst guaranteeing interpretability and transferability. In this work, we demonstrate the wide utility of prediction rigidities, a family of metrics derived from the loss function, in understanding the robustness of ML model predictions. We show that the prediction rigidities allow the assessment of the model not only at the global level, but also on the local or the component-wise level at which the intermediate (e.g. atomic, body-ordered, or range-separated) predictions are made. We leverage these metrics to understand the learning behavior of different ML models, and to guide efficient dataset construction for model training. We finally implement the formalism for a ML model targeting a coarse-grained system to demonstrate the applicability of the prediction rigidities to an even broader class of atomistic modeling problems.

Structural Determinants of Chirally Selective Transport of Amino Acids through the α-Hemolysin Protein Nanopores of Free-Standing Planar Lipid Membranes.

Despite the importance of the enantioselective transport of amino acids through transmembrane protein nanopores from fundamental and practical perspectives, little has been explored to date. Here, we study the transport of amino acids through α-hemolysin (αHL) protein pores incorporated into a free-standing lipid membrane. By measuring the transport of 13 different amino acids through the αHL pores, we discover that the molecular size of the amino acids and their capability to form hydrogen bonds with the pore surface determine the chiral selectivity. Molecular dynamics simulations corroborate our findings by revealing the enantioselective molecular-level interactions between the amino acid enantiomers and the αHL pore. Our work is the first to present the determinants for chiral selectivity using αHL protein as a molecular filter.

Robustness of Local Predictions in Atomistic Machine Learning Models.

Machine learning (ML) models for molecules and materials commonly rely on a decomposition of the global target quantity into local, atom-centered contributions. This approach is convenient from a computational perspective, enabling large-scale ML-driven simulations with a linear-scaling cost and also allows for the identification and posthoc interpretation of contributions from individual chemical environments and motifs to complicated macroscopic properties. However, even though practical justifications exist for the local decomposition, only the global quantity is rigorously defined. Thus, when the atom-centered contributions are used, their sensitivity to the training strategy or the model architecture should be carefully considered. To this end, we introduce a quantitative metric, which we call the local prediction rigidity (LPR), that allows one to assess how robust the locally decomposed predictions of ML models are. We investigate the dependence of the LPR on the aspects of model training, particularly the composition of training data set, for a range of different problems from simple toy models to real chemical systems. We present strategies to systematically enhance the LPR, which can be used to improve the robustness, interpretability, and transferability of atomistic ML models.

Flash-Thermal Shock Synthesis of Single Atoms in Ambient Air.

Single-atom catalysts feature interesting catalytic activity toward applications that rely on surface reactions such as electrochemical energy storage, catalysis, and gas sensors. However, conventional synthetic approaches for such catalysts require extended periods of high-temperature annealing in vacuum systems, limiting their throughput and increasing their production cost. Herein, we report an ultrafast flash-thermal shock (FTS)-induced annealing technique (temperature > 2850 °C, <10 ms duration, and ramping/cooling rates of ∼105 K/s) that operates in an ambient-air environment to prepare single-atom-stabilized N-doped graphene. Melamine is utilized as an N-doping source to provide thermodynamically favorable metal-nitrogen bonding sites, resulting in a uniform and high-density atomic distribution of single metal atoms. To demonstrate the practical utility of the single-atom-stabilized N-doped graphene produced by the FTS method, we showcased their chemiresistive gas sensing capabilities and electrocatalytic activities. Overall, the air-ambient, ultrafast, and versatile (e.g., Co, Ni, Pt, and Co-Ni dual metal) FTS method provides a general route for high-throughput, large area, and vacuum-free manufacturing of single-atom catalysts.

Rapid and Reversible Sensing Performance of Hydrogen-Substituted Graphdiyne.

The design of new nanomaterials for rapid and reversible detection of molecules in existence is critical for real-world sensing applications. Current nanomaterial libraries such as carbon nanotubes, graphene, MoS2, and MXene are fundamentally limited by their slow detection speed and small signals; thus, the atomic-level material design of molecular transport pathways and active binding sites must be accompanied. Herein, we fully explore the chemical and physical properties of a hydrogen-substituted graphdiyne (HsGDY) for its molecular sensing properties. This new carbon framework comprises reactive sp carbons in acetylenic linkages throughout the 16.3 Å nanopores and allows for detecting target molecules (e.g., H2) with an exceptionally high sensitivity (ΔR/Rb = 542%) and fast response/recovery time (τ90 = 8 s and τ10 = 38 s) even without any postmodification process. It possesses 2 orders of magnitude higher sensing ability than that of existing nanomaterial libraries. We demonstrate that rapid and reversible molecular binding is attributed to the cooperative interaction with adjacent double sp carbon in the layered nanoporous structure of HsGDY. This new class of carbon framework provides fundamental solutions for nanomaterials in reliable sensor applications that accelerate real-world interfacing.

Oxide/ZIF-8 Hybrid Nanofiber Yarns: Heightened Surface Activity for Exceptional Chemiresistive Sensing.

Though highly promising as powerful gas sensors, oxide semiconductor chemiresistors have low surface reactivity, which limits their selectivity, sensitivity, and reaction kinetics, particularly at room temperature (RT) operation. It is proposed that a hybrid design involving the nanostructuring of oxides and passivation with selective gas filtration layers can potentially overcome the issues with surface activity. Herein, unique bi-stacked heterogeneous layers are introduced; that is, nanostructured oxides covered by conformal nanoporous gas filters, on ultrahigh-density nanofiber (NF) yarns via sputter deposition with indium tin oxide (ITO) and subsequent self-assembly of zeolitic imidazolate framework (ZIF-8) nanocrystals. The NF yarn composed of ZIF-8-coated ITO films can offer heightened surface activity at RT because of high porosity, large surface area, and effective screening of interfering gases. As a case study, the hybrid sensor demonstrated remarkable sensing performances characterized by high NO selectivity, fast response/recovery kinetics (>60-fold improvement), and large responses (12.8-fold improvement @ 1 ppm) in comparison with pristine yarn@ITO, especially under highly humid conditions. Molecular modeling reveals an increased penetration ratio of NO over O2 to the ITO surface, indicating that NO oxidation is reliably prevented and that the secondary adsorption sites provided by the ZIF-8 facilitate the adsorption/desorption of NO, both to and from ITO.

Confinement of Ultrasmall Bimetallic Nanoparticles in Conductive Metal-Organic Frameworks via Site-Specific Nucleation.

Conductive metal-organic frameworks (cMOFs) are emerging materials for various applications due to their high surface area, high porosity, and electrical conductivity. However, it is still challenging to develop cMOFs having high surface reactivity and durability. Here, highly active and stable cMOF are presented via the confinement of bimetallic nanoparticles (BNPs) in the pores of a 2D cMOF, where the confinement is guided by dipolar-interaction-induced site-specific nucleation. Heterogeneous metal precursors are bound to the pores of 2D cMOFs by dipolar interactions, and the subsequent reduction produces ultrasmall (≈1.54 nm) and well-dispersed PtRu NPs confined in the pores of the cMOF. PtRu-NP-decorated cMOFs exhibit significantly enhanced chemiresistive NO2 sensing performances, owing to the bimetallic synergies of PtRu NPs and the high surface area and porosity of cMOF. The approach paves the way for the synthesis of highly active and conductive porous materials via bimetallic and/or multimetallic NP loading.

2D Materials Decorated with Ultrathin and Porous Graphene Oxide for High Stability and Selective Surface Activity.

2D black phosphorus (BP) and MXenes have triggered enormous research interest in catalysis, energy storage, and chemical sensing. Unfortunately, the low stability of these materials under practical operating conditions remains a critical bottleneck, particularly as they are prone to oxidization under moisture. In this work, the design and application of stable 2D heterostructures obtained from decorating BP and MXene (Ti3 C2 Tx ) with few-layer holey graphene oxide (FHGO) membranes are presented. In the resulting heterostructured systems, FHGO serves as a multifunctional passivation layer that shields BP or MXene from oxidative degradation, while allowing the selective diffusion of target gas molecules through its micropores and toward the underlying 2D material. Through a case study of dilute NO2 sensing, it is demonstrated that these heterostructures show a greatly enhanced sensing performance under humid conditions, where fast sensing speed and response are consistently observed, and high stability is impressively retained upon repetitive sensing cycles for 1000 min. These results corroborate the efficacy of material decoration with porous FHGO membranes and suggest that this is a generalizable strategy for reliable high-performance applications of 2D materials.

Rational modifications of PCN-700 to induce electrical conductivity: a computational study.

Recent research has shown that electrical conductivity can be observed in metal-organic frameworks (MOFs), which leads to many new application fields for these nanoporous materials. With a limited number of electrically conductive MOFs developed thus far, effective design strategies to induce electrical conductivity in these materials must be actively explored. In this computational study, we show that rational modifications of a previously insulating MOF, PCN-700, can lead to newfound electrical conductivity. In order to secure through-bond charge transport (CT) pathways in the framework, we consider the possibility of introducing electroactive DHBQ linkers via sequential linker installation. Then, metal substitution of Zr4+ with Ce4+ and saturation of DHBQ linkers at remaining linker vacant sites are additionally considered for the optimal matching of energy levels along the proposed CT pathway. The resulting linker saturated Ce-PCN-700-DHBQ is predicted to show semiconducting behavior with a bandgap of 2.09 eV, which can be further reduced by controlling the chemical environment. These computational results show that rational modifications of the framework can lead to electrical conductivity in previously insulating PCN-700, and highlight the importance of energy level matching in the design of electrically conductive MOFs.

Edge-Functionalized Graphene Nanoribbon Chemical Sensor: Comparison with Carbon Nanotube and Graphene.

With growing focus on the use of carbon nanomaterials in chemical sensors, one-dimensional graphene nanoribbon (GNR) has become one of the most attractive channel materials, owing to its enhanced conductance fluctuation by quantum confinement effects and dense, abundant edge sites. Due to the narrow width of a basal plane with one-dimensional morphology, chemical modification of edge sites would greatly affect the electrical channel properties of a GNR. Here, we demonstrate for the first time that chemically functionalizing the edge sites with aminopropylsilane (APS) molecules can significantly enhance the sensing performance of the GNR sensor. The resulting APS-functionalized GNR has a sensitivity ((Δ R/ Rb)max) of ∼30% at 0.125 ppm nitrogen dioxide (NO2) and an ultrafast response time (∼6 s), which are, respectively, 7- and 15-fold enhancements compared to a pristine GNR sensor. This is the fastest and most sensitive gas-sensing performance of all GNR sensors reported. To demonstrate the superiority of the GNR-APS sensor, we compare its sensing performance with that of APS-functionalized carbon nanotube (CNT) and reduced graphene oxide (rGO) sensors prepared in identical synthesis conditions. Very interestingly, the GNR-APS sensor exhibited 30- and 93-fold enhanced sensitivity compared to the CNT-APS and rGO-APS sensors. This might be attributed to highly active edge sites with superior chemical reactivity, which are not present in CNT and rGO materials. Density functional theory clearly shows that the greatly enhanced gas response of GNR with edge functionalization can be attributed to the higher electron densities in the highest occupied molecular orbital levels of GNR-APS and incorporation of additional adsorption sites. This finding is the first demonstration of the importance of edge functionalization of GNR for chemical sensors.

Rational Design of Aminopolymer for Selective Discrimination of Acidic Air Pollutants.

Strong acidic gases such as CO2, SO2, and NO2 are harsh air pollutants with major human health threatening factors, and as such, developing new tools to monitor and to quickly sense these gases is critically required. However, it is difficult to selectively detect the acidic air pollutants with single channel material due to the similar chemistry shared by acidic molecules. In this work, three acidic gases (i.e., CO2, SO2, and NO2) are selectively discriminated using single channel material with precise moiety design. By changing the composition ratio of primary (1°), secondary (2°), and tertiary (3°) amines of polyethylenimine (PEI) on CNT channels, unprecedented high selectivity between CO2 and SO2 is achieved. Using in situ FT-IR characterizations, the distinct adsorption phenomenon of acidic gases on each amine moiety is precisely demonstrated. Our approach is the first attempt at controlling gas adsorption selectivity of solid-state sensor via modulating chemical moiety level within the single channel material. In addition, discrimination of CO2, SO2, and NO2 with the single channel material solid-state sensor is first reported. We believe that this approach can greatly enhance air pollution tracking systems for strong acidic pollutants and thus aid future studies on selective solid-state gas sensors.

Excavating hidden adsorption sites in metal-organic frameworks using rational defect engineering.

Metal-organic frameworks are known to contain defects within their crystalline structures. Successful engineering of these defects can lead to modifications in material properties that can potentially improve the performance of many existing frameworks. Herein, we report the high-throughput computational screening of a large experimental metal-organic framework database to identify 13 frameworks that show significantly improved methane storage capacities with linker vacancy defects. The candidates are first identified by focusing on structures with methane-inaccessible pores blocked away from the main adsorption channels. Then, organic linkers of the candidate structures are judiciously replaced with appropriate modulators to emulate the presence of linker vacancies, resulting in the integration and utilization of the previously inaccessible pores. Grand canonical Monte Carlo simulations of defective candidate frameworks show significant enhancements in methane storage capacities, highlighting that rational defect engineering can be an effective method to significantly improve the performance of the existing metal-organic frameworks.