Cong Su

Transition-metal dichalcogenide (TMD) monolayers exhibit unique electronic, photonic, and quantum phenomena, yet their material quality remains constrained by defects and thickness inhomogeneity during chemical vapor deposition. Here, we identify the limitations of the common metal trioxide precursors: high volatility that induces stochastic vapor-phase nucleation and multilayer growth, and liberated oxygen-species-mediated chemical etchants that degrade lattice integrity. We demonstrate that an acid-mediated one-step modification, dissolving trioxides in hydrochloric acid, fundamentally redirects the precursor chemistry toward nonvolatile and substrate-anchored dioxide phase. This enforces a spatially confined solid-phase chalcogenization (SPC), minimizing the vapor-phase species and thereby suppressing dechalcogenization and vertical growth. The resulting uniform monolayers, synthesized as isolated triangular flakes or continuous films, achieve state-of-the-art low defect densities: 1.87 × 10 12 cm –2 for MoS 2 and 1.26 × 10 12 cm –2 for WSe 2 . Our work establishes SPC as a simple and unified mechanistic framework to drive TMD synthesis toward the intrinsic structural limits.

Fairness-aware machine learning aims to build predictive models that comply with fairness requirements, particularly concerning sensitive attributes such as race, gender, and age. Among causality-based fairness notions, counterfactual fairness is widely adopted for its individual-level guarantees, requiring that an individual’s predicted outcome remains unchanged in a counterfactual world where its sensitive attribute is altered. However, existing methods critically assume that the true causal graph is fully known, which is rarely the case in practice. Moreover, counterfactual fairness suffers from inherent identifiability limitations, as counterfactual quantities cannot always be uniquely estimated from observational data, especially under incomplete causal knowledge. To address these challenges, we propose a principled framework (CF-ICG) for counterfactual fairness under imperfectly known causal graphs, e.g., Completed Partially Directed Acyclic Graphs (CPDAGs). We first introduce a criterion to determine the identifiability, and bound the counterfactual quantities under CPDAGs. Building upon this, we develop an efficient local algorithm that avoids the exhaustive enumeration of all DAGs, ensuring robustness against worst-case fairness violations. Experimental results on synthetic and real-world datasets demonstrate the practical effectiveness and theoretical soundness of CF-ICG.

Background Angiogenesis and coronary collateral circulation (CCC) formation promote cardiac repair following severe coronary stenosis (SCS) or myocardial infarction (MI). Angiopoietin-like protein 7 (ANGPTL7) is a secreted protein associated with angiogenesis, but its role in CCC formation remains unclear. Objective The aim of this study was to investigate the role of ANGPTL7 in angiogenesis and evaluate the predictive value of serum ANGPTL7 in CCC formation and the prognosis of patients with SCS. Materials and methods The RNA sequencing was performed on myocardial tissues of mice to analyze the alterations of angiogenesis-related genes after MI. 100 patients with angiographically proven SCS and 36 controls were enrolled and retrospectively followed up. Serum ANGPTL7 was measured by enzyme-linked immunosorbent assays (ELISA). Human umbilical vein endothelial cells (HUVECs) and exogenous human recombinant ANGPTL7 protein were applied in assays including CCK-8, scratch, tube formation, cell immunofluorescence and western blot to demonstrate the proangiogenic effect of ANGPTL7. Gene Set Enrichment Analysis (GSEA) was used to perform KEGG pathway enrichment analysis on downstream mechanisms by which ANGPTL7 promoted angiogenesis. Results The transcriptional level of Angptl7 was upregulated in ischemic myocardial tissues of MI mice, and its serum levels increased in both mice post-MI and patients with SCS. Spearman correlation analysis indicated that serum ANGPTL7 levels were positively correlated to CCC grades ( r = 0.518, P < 0.001). Kaplan–Meier curves showed a higher serum ANGPTL7 was associated with a lower incidence of major adverse cardiovascular events (MACE) in patients with SCS (Log-rank test, P = 0.002). Cox proportional hazards regression analyses showed that serum ANGPTL7 level was remained a protective factor after adjusting for different covariates. Time-dependent receiver-operating characteristics (ROC) curves further explored the prognostic value of ANGPTL7, with the area under the curve (AUC) of 0.77 at 1 year, 0.70 at 2 years and 0.85 at 3 years. Additionally, ANGPTL7 enhanced endothelial cell proliferation, migration and capillary-like structure formation, indicating a proangiogenic effect in vitro . Conclusion ANGPTL7 serves as a predictive biomarker for CCC levels and the prognosis of patients with SCS, which probably attributed to its proangiogenic properties.

This study investigates the effects of embodied (breathing exercises) and cognitive interventions on adolescents’ flow experience and cognition patterns. Using a mixed-methods design, 303 vocational high school students were assigned to three groups: Embodied Task Group (N = 108), Cognitive Task Group (N = 100), and Mental Health Course Group (N = 95). Experiment 1 employed a 3×2 Multivariate Analysis of Covariance (MANCOVA) design to compare flow experience dimensions, while Experiment 2 used Epistemic Network Analysis (ENA) to analyze diary entries. Results showed that the Embodied Task Group outperformed the Cognitive Task Group in “Unambiguous Feedback” (ηp2 = 0.01, a small effect) and had higher “Transformation of Time” (ηp2 = 0.01, a small effect) than the Mental Health Course Group. ENA revealed that the Embodied Group developed stronger body-environment interaction patterns, shifting cognition pattern from psychological evaluations to dynamic bodily processes over time. Conversely, the Cognitive Task Group maintained event-focused cognition with weaker mind–body integration. Findings highlight breathing exercises’ potential to enhance flow experience through embodied awareness and multisensory processing, offering practical implications for mental health education by promoting embodied learning tasks to foster flow experience.

Excitons -- elementary excitations formed by bound electron-hole pairs -- govern the optical properties and excited-state dynamics of materials. In two-dimensions (2D), excitons are theoretically predicted to have a linear energy-momentum relation with a non-analytic discontinuity in the long wavelength limit, mimicking the dispersion of a photon. This results in an exciton that behaves like a massless particle, despite the fact that it is a composite boson composed of massive constituents. However, experimental observation of massless excitons has remained elusive. In this work, we unambiguously experimentally observe the predicted linear exciton dispersion in freestanding monolayer hexagonal boron nitride (hBN) using momentum-resolved electron energy-loss spectroscopy. The experimental result is in excellent agreement with our theoretical prediction based on ab initio many-body perturbation theory. Additionally, we identify the lowest dipole-allowed transition in monolayer hBN to be at 6.6 eV, illuminating a long-standing debate about the band gap of monolayer hBN. These findings provide critical insights into 2D excitonic physics and open new avenues for exciton-mediated superconductivity, Bose-Einstein condensation, and high-efficiency optoelectronic applications.

BACKGROUND: Observational studies have consistently reported positive associations between inflammatory biomarkers and the risk of developing aortic stenosis (AS). However, it is crucial to acknowledge that conventional observational studies are prone to various forms of bias, including reverse causation and residual confounding. To delve deeper into unraveling the potential causal relationship between inflammatory biomarkers and aortic stenosis, we conducted a comprehensive two-sample Mendelian randomization (MR) analysis. METHODS: In order to explore the causal effect of exposure to various circulating cytokines on the risk of developing AS, we carefully selected AS datasets as the exposures from the summary statistics of the genome-wide association study (GWAS) conducted by FinnGen. The dataset consisted of a sample size of 3283 for AS cases and 210,463 for controls. To estimate the MR analysis, we primarily adopted the inverse variance weighted (IVW) method. Additionally, we employed complementary methods, including Weighted Median, MR Egger, Weighted Mode, and Simple Mode, to analyze the causal associations comprehensively. In order to assess the presence of heterogeneity, we utilized Cochran's Q statistic and MR-Egger regression. To ensure the robustness and consistency of our findings, we conducted a leave-one-out analysis. RESULT: We observed a positive association between interleukin-18 (IL-18) levels and AS (odds ratio [OR] per standard deviation [SD] = 1.080; 95 % confidence interval [CI] 1.024 to 1.139), as well as between interferon-gamma levels (IFN-γ) and AS (OR per SD = 1.157; 95 % CI 1.028 to 1.302). Conversely, we found an inverse association between interleukin-13 (IL-13) levels and AS (OR per SD = 0.942; 95 % CI 0.890 to 0.997), as well as between interleukin-5 (IL-5) levels and AS (OR per SD = 0.892; 95 % CI 0.804 to 0.990). CONCLUSION: Our research enhances the current understanding of the role of specific inflammatory biomarker pathways in aortic stenosis. Nevertheless, further validation is required to assess the viability of targeting these cytokines through pharmacological or lifestyle interventions as potential treatments for aortic stenosis.

We have developed an efficient detector of secondary electrons (SEs) for a high-performance scanning transmission electron microscope (STEM) and tested it on several materials. Using the detector at 60 keV, we resolved the nearest neighbor atoms separated by 0.142 nm in SE images of graphene, and detected single-atom substitutions in graphene and monolayer MoS 2 . We imaged single heavy atoms on an amorphous carbon thin film, and the surface structure of gold nanoparticles supported on a thin film as well as on a bulk substrate. Other application examples shown in this paper include SE imaging combined with 4D STEM, simultaneous SE and electron energy loss spectroscopy (EELS) imaging, and simultaneous imaging of entrance and exit sides of a sample using two separate SE detectors. The results point to an exciting future for atomic-resolution SE imaging.

Causal learning is a recent and widely adopted paradigm to handle algorithmic discrimination. Contemporary causality-based studies on fairness only capture the unfair causal effect of a single-dimensional sensitive attribute (i.e., individual-dimension, like gender) on the decision. They neglect the socially constructed nature of individual attributes, such as macro-dimensional factors. However, social science research shows that discrimination against an individual may be related to disadvantaged treatments, which operate at the macro-dimension (e.g., neighborhood economic level). This multi-dimensional conceptualization is pertinent to matters of fairness, and it is crucial to be fair for individuals across multiple dimensions. The hidden confounder is another bottleneck for addressing fairness concerns based on causal techniques. To tackle these issues, we present an approach, called MultiCFL, which accounts for multi-dimensional sources of discrimination and unifies them via causal tools. To handle hidden confounders, MultiCFL first trains a causal effect variational autoencoder as the causal estimator to learn the causal mechanisms behind observational data. Subsequently, it makes selective use of estimated causal relationships to construct a predictive model with multi-dimensional fairness. Experimental results confirm the effectiveness of MultiCFL, and prove the necessity of considering multiple dimensional properties to mitigate unfairness.

Systemic sclerosis (SSc) is an autoimmune connective tissue disease with skin fibrosis being the first and most common manifestation. Patients with SSc have a higher risk of developing malignant tumors than the general population. However, the sequence and underlying mechanisms linking SSc to malignancy remain controversial. This article presents the case of a 59-year-old woman who was diagnosed with SSc after developing skin fibrosis during neoadjuvant therapy for breast cancer. Despite aggressive antitumor treatments, including targeted therapy, SSc did not improve and progressed rapidly with increasing dermatofibrosis. Remarkably, the SSc entered remission following the cessation of antitumor therapy. Additionally, we reviewed the literature on SSc and malignant tumors, examined their relationship, and discussed key points regarding their identification and potential for adverse drug reactions.

To achieve accurate molecular diagnosis and early-stage intervention of disease on demand, there is an urgent need for the monitoring of multiple biomarkers and multipath information acquisition in living cells. The DNA combinatorial logic gate is an appropriate strategy for providing a systematic proof of concept with comprehensive information and function. Herein, a modular DNA logic gate nanomachine is designed for sufficient multistep reciprocal cell identification and therapy via the iteration of simple logic operations. In this logic gate system, this main module is constructed by G-quadruplex-locked gold nanocages (AuNCs), serving dual functions of drug encapsulation and cell recognition. The logic system is composed of OR, XNOR, AND, and NOR gates employing two intracellular disease biomarkers (microRNA 21 and microRNA 155) as inputs and the fluorescence signal of doxorubicin (Dox) as an output. The output signals of the four logic gates are iterated to process the imaging analysis data from the complex matrix in the living cell. Via positive and negative reciprocal feedback, the series circuit of different gates enables different functions, including the preliminary screening and the distinction of the cell type. Through the mutual preliminary screening and further proof, this logic system achieves accurate identification of cells, controlled drug release, and photothermal treatment using the AuNC as a photothermal transducer. This DNA logic system broadens the applications of the biocomputing system in disease screening and logic-controlled treatment fields.

Although the “obesity paradox” is comprehensively elucidated in heart failure (HF) with reduced ejection fraction (HFrEF) and HF with preserved ejection fraction (HFpEF), the role of body composition in left ventricular (LV) remodeling, LV reverse remodeling (LVRR), and clinical outcomes is still unclear for HF with mildly reduced ejection fraction (HFmrEF). Our study is a single-centre, prospective, and echocardiography-based study. Consecutive HFmrEF patients, defined as HF patients with a left ventricular ejection fraction (LVEF) between 40 and 49%, between January 2016 to December 2021 were included. Echocardiography was re-examined at 3-, 6-, and 12-month follow-up to assess the LVRR dynamically. Body mass index (BMI), fat mass, fat-free mass, percent body fat (PBF), CUN-BAE index, and lean mass index (LMI) were adopted as anthropometric parameters in our study to assess body composition. The primary outcome was LVRR, defined as: (1) a reduction higher than 10% in LV end-diastolic diameter index (LVEDDI), or a LVEDDI < 33 mm/m2, (2) an absolute increase of LVEF higher than 10 points compared with baseline echocardiogram, or a follow-up LVEF ≥50%. The secondary outcome was a composite of re-hospitalization for HF or cardiovascular death. A total of 240 HFmrEF patients were enrolled in our formal analysis. After 1-year follow-up based on echocardiography, 113 (47.1%) patients developed LVRR. Patients with LVRR had higher fat mass (21.7 kg vs. 19.3 kg, P = 0.034) and PBF (28.7% vs. 26.6%, P = 0.047) compared with those without. The negative correlation between anthropometric parameters and baseline LVEDDI was significant (all P < 0.05). HFmrEF patients with higher BMI, fat mass, PBF, CUN-BAE index, and LMI had more pronounced and persistent increase of LVEF and decline in LV mass index (LVMI). Univariable Cox regression analysis revealed that higher BMI (HR 1.042, 95% CI 1.002–1.083, P = 0.037) and fat mass (HR 1.019, 95% CI 1.002–1.036, P = 0.026) were each significantly associated with higher cumulative incidence of LVRR for HFmrEF patients, while this relationship vanished in the adjusted model. Mediation analysis indicated that the association between BMI and fat mass with LVRR was fully mediated by baseline LV dilation. Furthermore, higher fat mass (aHR 0.957, 95% CI 0.917–0.999, P = 0.049) and PBF (aHR 0.963, 95% CI 0.924–0.976, P = 0.043) was independently associated with lower risk of adverse clinical events. Body composition played an important role in the LVRR and clinical outcomes for HFmrEF. For HFmrEF patients, BMI and fat mass was positively associated with the cumulative incidence of LVRR, while higher fat mass and PBF predicted lower risk of adverse clinical events but not LMI.

We report on the three-dimensional (3D) structure determination of a twisted hexagonal boron nitride (h-BN) heterointerface from a single-view dataset using multislice ptychography. We identify the buried heterointerface between two twisted h-BN flakes with a lateral resolution of 0.57 \AA{} and a depth resolution of 2.5 nm. The latter represents a significant improvement (\ensuremath{\sim}2.7 times) over the aperture-limited depth resolution of incoherent imaging modes, such as annular-dark-field scanning transmission electron microscopy. This improvement is attributed to the diffraction signal extending beyond the aperture edge, with the depth resolution set by the curvature of the Ewald sphere. Future advancements in this approach could enhance the depth resolution to the subnanometer level and enable the identification of individual dopants, defects, and color centers in twisted heterointerfaces and other materials.

BACKGROUND: Serum uric acid (SUA) is an important pathogenetic and prognostic factor for heart failure (HF). Gender differences are apparent in HF. Furthermore, gender differences also exist in the association between SUA and prognosis in various cardiovascular diseases. However, the gender difference for SUA in the prediction of long-term prognosis in HF is still ambiguous. METHODS: A total of 1593 HF patients (897 men, 696 women) from the National Health and Nutrition Examination Survey (NHANES) 1999-2018 cycle were enrolled in our final analysis. Participants were categorized according to gender-specific SUA tertile. We assessed the association between SUA and long-term prognosis of HF patients, defined as all-cause mortality and cardiovascular mortality, in different genders via Kaplan-Meier curve analysis, Cox proportional hazard model, and Fine-Gray competing risk model. The restricted cubic spline (RCS) was performed to investigate the dose-response relationship between SUA and outcomes. RESULTS: Gender differences exist in demographic characteristics, clinical parameters, laboratory tests, and medication of HF patients. After a median follow-up of 127 months (95% CI 120-134 months), there were 853 all-cause deaths (493 events in men, 360 events in women) and 361 cardiovascular deaths (206 events in men, 155 events in women). Kaplan-Meier analysis showed that SUA had gender difference in the prediction of cardiovascular mortality (Log-rank p < 0.001, for male, Log-rank p = 0.150, for female), but not in all-cause mortality. Multivariate Cox regression analysis revealed that elevated SUA levels were associated with higher all-cause mortality and cardiovascular mortality in men (HR 1.11, 95% CI 1.05-1.18, p < 0.001, for all-cause death; HR 1.18, 95% CI 1.09-1.28, p < 0.001, for cardiovascular death), but not in women (HR 1.05, 95% CI 0.98-1.12, p = 0.186, for all-cause death; HR 1.01, 95% CI 0.91-1.12, p = 0.902, for cardiovascular death). Even using non-cardiovascular death as a competitive risk, adjusted Fine-Gray model also illustrated that SUA was an independent predictor of cardiovascular death in men (SHR 1.17, 95% CI 1.08-1.27, p < 0.001), but not in women (SHR 0.98, 95% CI 0.87 - 1.10, p = 0.690). CONCLUSIONS: Gender differences in the association between SUA and long-term prognosis of HF existed. SUA was an independent prognostic predictor for long-term outcomes of HF in men, but not in women.

Federated learning (FL) has emerged as a promising collaborative and secure paradigm for training a model from decentralized data without compromising privacy. Group fairness and client fairness are two dimensions of fairness that are important for FL. Standard FL can result in disproportionate disadvantages for certain clients, and it still faces the challenge of treating different groups equitably in a population. The problem of privately training fair FL models without compromising the generalization capability of disadvantaged clients remains open. In this paper, we propose a method, called mFairFL, to address this problem and achieve group fairness and client fairness simultaneously. mFairFL leverages differential multipliers to construct an optimization objective for empirical risk minimization with fairness constraints. Before aggregating locally trained models, it first detects conflicts among their gradients, and then iteratively curates the direction and magnitude of gradients to mitigate these conflicts. Theoretical analysis proves mFairFL facilitates the fairness in model development. The experimental evaluations based on three benchmark datasets show significant advantages of mFairFL compared to seven state-of-the-art baselines.

Hexagonal boron nitride (hBN), also known as white graphite, is a transparent layered crystal with a wide bandgap. Its crystal structure resembles graphite, featuring layers composed of honeycomb lattices held together through van der Waals forces. The layered crystal structure of hBN facilitates exfoliation into thinner flakes and makes it highly anisotropic in in-plane and out-of-plane directions. Unlike graphite, hBN is both insulating and transparent, making it an ideal material for isolating devices from the environment and acting as a waveguide. As a result, hBN has found extensive applications in optical devices, electronic devices, and quantum photonic devices. This comprehensive tutorial aims to provide readers with a thorough understanding of hBN, covering its synthesis, lattice and spectroscopic characterization, and various applications in optoelectronic and quantum photonic devices. This tutorial is designed for both readers without prior experience in hBN and those with expertise in specific fields seeking to understand its relevance and connections to others.

Background: At present, there is a paucity of research on the link between Crohn’s disease (CD) and atrial fibrillation (AF). Nevertheless, both ailments are thought to entail inflammatory and autoimmune processes, and emerging evidence indicates that individuals with CD may face an elevated risk of AF. To shed light on this issue, our study seeks to explore the possibility of shared genes, pathways, and immune cells between these two conditions. Methods: We retrieved the gene expression profiles of both CD and AF from the Gene Expression Omnibus (GEO) database and subjected them to analysis. Afterward, we utilized the weighted gene co-expression network analysis (WGCNA) to identify shared genes, which were then subjected to further Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses. Furthermore, we employed a rigorous analytical approach by screening hub genes through both least absolute shrinkage and selection operator (LASSO) regression and support vector machine (SVM), and subsequently constructing a receiver operating characteristic (ROC) curve based on the screening outcomes. Finally, we utilized single-sample gene set enrichment analysis (ssGSEA) to comprehensively evaluate the levels of infiltration of 28 immune cells within the expression profile and their potential association with the shared hub genes. Results: Using the WGCNA method, we identified 30 genes that appear to be involved in the pathological progression of both AF and CD. Through GO enrichment analysis on the key gene modules derived from WGCNA, we observed a significant enrichment of pathways related to major histocompatibility complex (MHC) and antigen processing. By leveraging the intersection of LASSO and SVM algorithms, we were able to pinpoint two overlapping genes, namely CXCL16 and HLA-DPB1. Additionally, we evaluated the infiltration of immune cells and observed the upregulation of CD4+ and CD8+ T cells, as well as dendritic cells in patients with AF and CD. Conclusions: By employing bioinformatics tools, we conducted an investigation with the objective of elucidating the genetic foundations that connect AF and CD. This study culminated in the identification of CXCL16 and HLA-DPB1 as the most substantial genes implicated in the development of both disorders. Our findings suggest that the immune responses mediated by CD4+ and CD8+ T cells, along with dendritic cells, may hold a crucial role in the intricate interplay between AF and CD.

Federated learning (FL) has emerged as a promising collaborative and secure paradigm for training a model from decentralized data without compromising privacy. Group fairness and client fairness are two dimensions of fairness that are important for FL. Standard FL can result in disproportionate disadvantages for certain clients, and it still faces the challenge of treating different groups equitably in a population. The problem of privately training fair FL models without compromising the generalization capability of disadvantaged clients remains open. In this paper, we propose a method, called mFairFL, to address this problem and achieve group fairness and client fairness simultaneously. mFairFL leverages differential multipliers to construct an optimization objective for empirical risk minimization with fairness constraints. Before aggregating locally trained models, it first detects conflicts among their gradients, and then iteratively curates the direction and magnitude of gradients to mitigate these conflicts. Theoretical analysis proves mFairFL facilitates the fairness in model development. The experimental evaluations based on three benchmark datasets show significant advantages of mFairFL compared to seven state-of-the-art baselines.

We report three-dimensional (3D) structure determination of a twisted hexagonal boron nitride (h-BN) heterointerface from a single-view data set using multislice ptychography. We identify the buried heterointerface between two twisted h-BN flakes with a lateral resolution of 0.57 Å and a depth resolution of 2.5 nm. The latter is a significant improvement (~2.7 times) over the aperture-limited depth resolution of incoherent imaging modes such as annular-dark-field scanning transmission electron microscopy. This is attributed to the diffraction signal extending beyond the aperture edge with the depth resolution set by the curvature of the Ewald sphere. Future advances to this approach could improve the depth resolution to the sub-nanometer level and enable the identification of individual dopants, defects and color centers in twisted heterointerfaces and other materials.

Abstract Four-dimensional scanning transmission electron microscopy (4D-STEM) has recently gained widespread attention for its ability to image atomic electric fields with sub-Ångstrom spatial resolution. These electric field maps represent the integrated effect of the nucleus, core electrons and valence electrons, and separating their contributions is non-trivial. In this paper, we utilized simultaneously acquired 4D-STEM center of mass (CoM) images and annular dark field (ADF) images to determine the projected electron charge density in monolayer MoS 2 . We evaluate the contributions of both the core electrons and the valence electrons to the derived electron charge density; however, due to blurring by the probe shape, the valence electron contribution forms a nearly featureless background while most of the spatial modulation comes from the core electrons. Our findings highlight the importance of probe shape in interpreting charge densities derived from 4D-STEM and the need for smaller electron probes.

The engineering of atomically-precise nanopores in two-dimensional materials presents exciting opportunities for both fundamental science studies as well as applications in energy, DNA sequencing, and quantum information technologies. The exceptional chemical and thermal stability of hexagonal boron nitride (h-BN) suggest that exposed h-BN nanopores will retain their atomic structure even when subjected to extended periods of time in gas or liquid environments. Here we employ transmission electron microscopy to examine the time evolution of h-BN nanopores in vacuum and in air and find, even at room temperature, dramatic geometry changes due to atom motion and edge contamination adsorption, for timescales ranging from one hour to one week. The discovery of nanopore evolution contrasts with general expectations and has profound implications for nanopore applications of two-dimensional materials.

We report the development of deep-learning coherent electron diffractive imaging at subangstrom resolution using convolutional neural networks (CNNs) trained with only simulated data. We experimentally demonstrate this method by applying the trained CNNs to recover the phase images from electron diffraction patterns of twisted hexagonal boron nitride, monolayer graphene, and a gold nanoparticle with comparable quality to those reconstructed by a conventional ptychographic algorithm. Fourier ring correlation between the CNN and ptychographic images indicates the achievement of a resolution in the range of 0.70 and 0.55 Å. We further develop CNNs to recover the probe function from the experimental data. The ability to replace iterative algorithms with CNNs and perform real-time atomic imaging from coherent diffraction patterns is expected to find applications in the physical and biological sciences.