Low-frequency stimulation enhances ensemble co-firing and dexterity after stroke.

Electrical stimulation is a promising tool for modulating brain networks. However, it is unclear how stimulation interacts with neural patterns underlying behavior. Specifically, how might external stimulation that is not sensitive to the state of ongoing neural dynamics reliably augment neural processing and improve function? Here, we tested how low-frequency epidural alternating current stimulation (ACS) in non-human primates recovering from stroke interacted with task-related activity in perilesional cortex and affected grasping. We found that ACS increased co-firing within task-related ensembles and improved dexterity. Using a neural network model, we found that simulated ACS drove ensemble co-firing and enhanced propagation of neural activity through parts of the network with impaired connectivity, suggesting a mechanism to link increased co-firing to enhanced dexterity. Together, our results demonstrate that ACS restores neural processing in impaired networks and improves dexterity following stroke. More broadly, these results demonstrate approaches to optimize stimulation to target neural dynamics.

Automating the analysis of facial emotion expression dynamics: A computational framework and application in psychotic disorders.

Facial emotion expressions play a central role in interpersonal interactions; these displays are used to predict and influence the behavior of others. Despite their importance, quantifying and analyzing the dynamics of brief facial emotion expressions remains an understudied methodological challenge. Here, we present a method that leverages machine learning and network modeling to assess the dynamics of facial expressions. Using video recordings of clinical interviews, we demonstrate the utility of this approach in a sample of 96 people diagnosed with psychotic disorders and 116 never-psychotic adults. Participants diagnosed with schizophrenia tended to move from neutral expressions to uncommon expressions (e.g., fear, surprise), whereas participants diagnosed with other psychoses (e.g., mood disorders with psychosis) moved toward expressions of sadness. This method has broad applications to the study of normal and altered expressions of emotion and can be integrated with telemedicine to improve psychiatric assessment and treatment.

PPICT: an integrated deep neural network for predicting inter-protein PTM cross-talk.

Post-translational modifications (PTMs) fine-tune various signaling pathways not only by the modification of a single residue, but also by the interplay of different modifications on residue pairs within or between proteins, defined as PTM cross-talk. As a challenging question, less attention has been given to PTM dynamics underlying cross-talk residue pairs and structural information underlying protein-protein interaction (PPI) graph, limiting the progress in this PTM functional research. Here we propose a novel integrated deep neural network PPICT (Predictor for PTM Inter-protein Cross-Talk), which predicts PTM cross-talk by combining protein sequence-structure-dynamics information and structural information for PPI graph. We find that cross-talk events preferentially occur among residues with high co-evolution and high potential in allosteric regulation. To make full use of the complex associations between protein evolutionary and biophysical features, and protein pair features, a heterogeneous feature combination net is introduced in the final prediction of PPICT. The comprehensive test results show that the proposed PPICT method significantly improves the prediction performance with an AUC value of 0.869, outperforming the existing state-of-the-art methods. Additionally, the PPICT method can capture the potential PTM cross-talks involved in the functional regulatory PTMs on modifying enzymes and their catalyzed PTM substrates. Therefore, PPICT represents an effective tool for identifying PTM cross-talk between proteins at the proteome level and highlights the hints for cross-talk between different signal pathways introduced by PTMs.

A circuit mechanism for independent modulation of excitatory and inhibitory firing rates after sensory deprivation.

Diverse interneuron subtypes shape sensory processing in mature cortical circuits. During development, sensory deprivation evokes powerful synaptic plasticity that alters circuitry, but how different inhibitory subtypes modulate circuit dynamics in response to this plasticity remains unclear. We investigate how deprivation-induced synaptic changes affect excitatory and inhibitory firing rates in a microcircuit model of the sensory cortex with multiple interneuron subtypes. We find that with a single interneuron subtype (parvalbumin-expressing [PV]), excitatory and inhibitory firing rates can only be comodulated-increased or decreased together. To explain the experimentally observed independent modulation, whereby one firing rate increases and the other decreases, requires strong feedback from a second interneuron subtype (somatostatin-expressing [SST]). Our model applies to the visual and somatosensory cortex, suggesting a general mechanism across sensory cortices. Therefore, we provide a mechanistic explanation for the differential role of interneuron subtypes in regulating firing rates, contributing to the already diverse roles they serve in the cortex.

A model of autonomous interactions between hippocampus and neocortex driving sleep-dependent memory consolidation.

How do we build up our knowledge of the world over time? Many theories of memory formation and consolidation have posited that the hippocampus stores new information, then "teaches" this information to the neocortex over time, especially during sleep. But it is unclear, mechanistically, how this actually works-How are these systems able to interact during periods with virtually no environmental input to accomplish useful learning and shifts in representation? We provide a framework for thinking about this question, with neural network model simulations serving as demonstrations. The model is composed of hippocampus and neocortical areas, which replay memories and interact with one another completely autonomously during simulated sleep. Oscillations are leveraged to support error-driven learning that leads to useful changes in memory representation and behavior. The model has a non-rapid eye movement (NREM) sleep stage, where dynamics between the hippocampus and neocortex are tightly coupled, with the hippocampus helping neocortex to reinstate high-fidelity versions of new attractors, and a REM sleep stage, where neocortex is able to more freely explore existing attractors. We find that alternating between NREM and REM sleep stages, which alternately focuses the model's replay on recent and remote information, facilitates graceful continual learning. We thus provide an account of how the hippocampus and neocortex can interact without any external input during sleep to drive useful new cortical learning and to protect old knowledge as new information is integrated.

Multielectrode Cortical Stimulation Selectively Induces Unidirectional Wave Propagation of Excitatory Neuronal Activity in Biophysical Neural Model.

Cortical stimulation is emerging as an experimental tool in basic research and a promising therapy for a range of neuropsychiatric conditions. As multielectrode arrays enter clinical practice, the possibility of using spatiotemporal patterns of electrical stimulation to induce desired physiological patterns has become theoretically possible, but in practice can only be implemented by trial-and-error because of a lack of predictive models. Experimental evidence increasingly establishes traveling waves as fundamental to cortical information-processing, but we lack an understanding of how to control wave properties despite rapidly improving technologies. This study uses a hybrid biophysical-anatomical and neural-computational model to predict and understand how a simple pattern of cortical surface stimulation could induce directional traveling waves via asymmetric activation of inhibitory interneurons. We found that pyramidal cells and basket cells are highly activated by the anodal electrode and minimally activated by the cathodal electrodes, while Martinotti cells are moderately activated by both electrodes but exhibit a slight preference for cathodal stimulation. Network model simulations found that this asymmetrical activation results in a traveling wave in superficial excitatory cells that propagates unidirectionally away from the electrode array. Our study reveals how asymmetric electrical stimulation can easily facilitate traveling waves by relying on two distinct types of inhibitory interneuron activity to shape and sustain the spatiotemporal dynamics of endogenous local circuit mechanisms.SIGNIFICANCE STATEMENT Electrical brain stimulation is becoming increasingly useful to probe the workings of brain and to treat a variety of neuropsychiatric disorders. However, stimulation is currently performed in a trial-and-error fashion as there are no methods to predict how different electrode arrangements and stimulation paradigms will affect brain functioning. In this study, we demonstrate a hybrid modeling approach, which makes experimentally testable predictions that bridge the gap between the microscale effects of multielectrode stimulation and the resultant circuit dynamics at the mesoscale. Our results show how custom stimulation paradigms can induce predictable, persistent changes in brain activity, which has the potential to restore normal brain function and become a powerful therapy for neurological and psychiatric conditions.

Dynamically driven correlations in elastic net models reveal sequence of events and causality in proteins.

An explicit analytic solution is given for the Langevin equation applied to the Gaussian Network Model of a protein subjected to both a random and a deterministic periodic force. Synchronous and asynchronous components of time correlation functions are derived and an expression for phase differences in the time correlations of residue pairs is obtained. The synchronous component enables the determination of dynamic communities within the protein structure. The asynchronous component reveals causality, where the time correlation function between residues i and j differs depending on whether i is observed before j or vice versa, resulting in directional information flow. Driver and driven residues in the allosteric process of cyclophilin A and human NAD-dependent isocitrate dehydrogenase are determined by a perturbation-scanning technique. Factors affecting phase differences between fluctuations of residues, such as network topology, connectivity, and residue centrality, are identified. Within the constraints of the isotropic Gaussian Network Model, our results show that asynchronicity increases with viscosity and distance between residues, decreases with increasing connectivity, and decreases with increasing levels of eigenvector centrality.

Genome-scale metabolic network model and phenome of solvent-tolerant Pseudomonas putida S12.

BACKGROUND: Pseudomonas putida S12 is a gram-negative bacterium renowned for its high tolerance to organic solvents and metabolic versatility, making it attractive for various applications, including bioremediation and the production of aromatic compounds, bioplastics, biofuels, and value-added compounds. However, a metabolic model of S12 has yet to be developed. RESULTS: In this study, we present a comprehensive and highly curated genome-scale metabolic network model of S12 (iSH1474), containing 1,474 genes, 1,436 unique metabolites, and 2,938 metabolic reactions. The model was constructed by leveraging existing metabolic models and conducting comparative analyses of genomes and phenomes. Approximately 2,000 different phenotypes were measured for S12 and its closely related KT2440 strain under various nutritional and environmental conditions. These phenotypic data, combined with the reported experimental data, were used to refine and validate the reconstruction. Model predictions quantitatively agreed well with in vivo flux measurements and the batch cultivation of S12, which demonstrated that iSH1474 accurately represents the metabolic capabilities of S12. Furthermore, the model was simulated to investigate the maximum theoretical metabolic capacity of S12 growing on toxic organic solvents. CONCLUSIONS: iSH1474 represents a significant advancement in our understanding of the cellular metabolism of P. putida S12. The combined results of metabolic simulation and comparative genome and phenome analyses identified the genetic and metabolic determinants of the characteristic phenotypes of S12. This study could accelerate the development of this versatile organism as an efficient cell factory for various biotechnological applications.

Differential sensitivity to longitudinal and transverse stretch mediates transcriptional responses in mouse neonatal ventricular myocytes.

To identify how cardiomyocyte mechanosensitive signaling pathways are regulated by anisotropic stretch, micropatterned mouse neonatal cardiomyocytes were stretched primarily longitudinally or transversely to the myofiber axis. Four hours of static, longitudinal stretch induced differential expression of 557 genes, compared with 30 induced by transverse stretch, measured using RNA-seq. A logic-based ordinary differential equation model of the cardiac myocyte mechanosignaling network, extended to include the transcriptional regulation and expression of 784 genes, correctly predicted measured expression changes due to anisotropic stretch with 69% accuracy. The model also predicted published transcriptional responses to mechanical load in vitro or in vivo with 63-91% accuracy. The observed differences between transverse and longitudinal stretch responses were not explained by differential activation of specific pathways but rather by an approximately twofold greater sensitivity to longitudinal stretch than transverse stretch. In vitro experiments confirmed model predictions that stretch-induced gene expression is more sensitive to angiotensin II and endothelin-1, via RhoA and MAP kinases, than to the three membrane ion channels upstream of calcium signaling in the network. Quantitative cardiomyocyte gene expression differs substantially with the axis of maximum principal stretch relative to the myofilament axis, but this difference is due primarily to differences in stretch sensitivity rather than to selective activation of mechanosignaling pathways.NEW & NOTEWORTHY Anisotropic stretch applied to micropatterned neonatal mouse ventricular myocytes induced markedly greater acute transcriptional responses when the major axis of stretch was parallel to the myofilament axis than when it was transverse. Analysis with a novel quantitative network model of mechanoregulated cardiomyocyte gene expression suggests that this difference is explained by higher cell sensitivity to longitudinal loading than transverse loading than by the activation of differential signaling pathways.

Self-controlled in silico gene knockdown strategies to enhance the sustainable production of heterologous terpenoid by Saccharomyces cerevisiae.

Microbial bioengineering is a growing field for producing plant natural products (PNPs) in recent decades, using heterologous metabolic pathways in host cells. Once heterologous metabolic pathways have been introduced into host cells, traditional metabolic engineering techniques are employed to enhance the productivity and yield of PNP biosynthetic routes, as well as to manage competing pathways. The advent of computational biology has marked the beginning of a novel epoch in strain design through in silico methods. These methods utilize genome-scale metabolic models (GEMs) and flux optimization algorithms to facilitate rational design across the entire cellular metabolic network. However, the implementation of in silico strategies can often result in an uneven distribution of metabolic fluxes due to the rigid knocking out of endogenous genes, which can impede cell growth and ultimately impact the accumulation of target products. In this study, we creatively utilized synthetic biology to refine in silico strain design for efficient PNPs production. OptKnock simulation was performed on the GEM of Saccharomyces cerevisiae OA07, an engineered strain for oleanolic acid (OA) bioproduction that has been reported previously. The simulation predicted that the single deletion of fol1, fol2, fol3, abz1, and abz2, or a combined knockout of hfd1, ald2 and ald3 could improve its OA production. Consequently, strains EK1∼EK7 were constructed and cultivated. EK3 (OA07△fol3), EK5 (OA07△abz1), and EK6 (OA07△abz2) had significantly higher OA titers in a batch cultivation compared to the original strain OA07. However, these increases were less pronounced in the fed-batch mode, indicating that gene deletion did not support sustainable OA production. To address this, we designed a negative feedback circuit regulated by malonyl-CoA, a growth-associated intermediate whose synthesis served as a bypass to OA synthesis, at fol3, abz1, abz2, and at acetyl-CoA carboxylase-encoding gene acc1, to dynamically and autonomously regulate the expression of these genes in OA07. The constructed strains R_3A, R_5A and R_6A had significantly higher OA titers than the initial strain and the responding gene-knockout mutants in either batch or fed-batch culture modes. Among them, strain R_3A stand out with the highest OA titer reported to date. Its OA titer doubled that of the initial strain in the flask-level fed-batch cultivation, and achieved at 1.23 ± 0.04 g L-1 in 96 h in the fermenter-level fed-batch mode. This indicated that the integration of optimization algorithm and synthetic biology approaches was efficiently rational for PNP-producing strain design.

Systems-level transcriptional regulation of Caenorhabditis elegans metabolism.

Metabolism is controlled to ensure organismal development and homeostasis. Several mechanisms regulate metabolism, including allosteric control and transcriptional regulation of metabolic enzymes and transporters. So far, metabolism regulation has mostly been described for individual genes and pathways, and the extent of transcriptional regulation of the entire metabolic network remains largely unknown. Here, we find that three-quarters of all metabolic genes are transcriptionally regulated in the nematode Caenorhabditis elegans. We find that many annotated metabolic pathways are coexpressed, and we use gene expression data and the iCEL1314 metabolic network model to define coregulated subpathways in an unbiased manner. Using a large gene expression compendium, we determine the conditions where subpathways exhibit strong coexpression. Finally, we develop "WormClust," a web application that enables a gene-by-gene query of genes to view their association with metabolic (sub)-pathways. Overall, this study sheds light on the ubiquity of transcriptional regulation of metabolism and provides a blueprint for similar studies in other organisms, including humans.

Tunnel-like region observed as a potential allosteric site in Staphylococcus aureus Glyceraldehyde-3-phosphate dehydrogenase.

Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) catalyzing the sixth step of glycolysis has been investigated for allosteric features that might be used as potential target for specific inhibition of Staphylococcus aureus (S.aureus). X-ray structure of bacterial enzyme for which a tunnel-like opening passing through the center previously proposed as an allosteric site has been subjected to six independent 500 ns long Molecular Dynamics simulations. Harmonic bond restraints were employed at key residues to underline the allosteric feature of this region. A noticeable reduction was observed in the mobility of NAD+ binding domains when restrictions were applied. Also, a substantial decrease in cross-correlations between distant Cα fluctuations was detected throughout the structure. Mutual information (MI) analysis revealed a similar decrease in the degree of correspondence in positional fluctuations in all directions everywhere in the receptor. MI between backbone and side-chain torsional variations changed its distribution profile and decreased considerably around the catalytic sites when restraints were employed. Principal component analysis clearly showed that the restrained state sampled a narrower range of conformations than apo state, especially in the first principal mode due to restriction in the conformational flexibility of NAD+ binding domain. Clustering the trajectory based on catalytic site residues displayed a smaller repertoire of conformations for restrained state compared to apo. Representative snapshots subjected to k-shortest pathway analysis revealed the impact of bond restraints on the allosteric communication which displayed distinct optimal and suboptimal pathways for two states, where observed frequencies of critical residues Gln51 and Val283 at the proposed site changed considerably.

A mean-field model of gamma-frequency oscillations in networks of excitatory and inhibitory neurons.

Gamma oscillations are widely seen in the cerebral cortex in different states of the wake-sleep cycle and are thought to play a role in sensory processing and cognition. Here, we study the emergence of gamma oscillations at two levels, in networks of spiking neurons, and a mean-field model. At the network level, we consider two different mechanisms to generate gamma oscillations and show that they are best seen if one takes into account the synaptic delay between neurons. At the mean-field level, we show that, by introducing delays, the mean-field can also produce gamma oscillations. The mean-field matches the mean activity of excitatory and inhibitory populations of the spiking network, as well as their oscillation frequencies, for both mechanisms. This mean-field model of gamma oscillations should be a useful tool to investigate large-scale interactions through gamma oscillations in the brain.

Development and applications of genome-scale metabolic network models.

The genome-scale metabolic network model is an effective tool for characterizing the gene-protein-response relationship in the entire metabolic pathway of an organism. By combining various algorithms, the genome-scale metabolic network model can effectively simulate the influence of a specific environment on the physiological state of cells, optimize the culture conditions of strains, and predict the targets of genetic modification to achieve targeted modification of strains. In this review, we summarize the whole process of model building, sort out the various tools that may be involved in the model building process, and explain the role of various algorithms in model analysis. In addition, we also summarized the application of GSMM in network characteristics, cell phenotypes, metabolic engineering, etc. Finally, we discuss the current challenges facing GSMM.

Accounting for network noise in graph-guided Bayesian modeling of structured high-dimensional data.

There is a growing body of literature on knowledge-guided statistical learning methods for analysis of structured high-dimensional data (such as genomic and transcriptomic data) that can incorporate knowledge of underlying networks derived from functional genomics and functional proteomics. These methods have been shown to improve variable selection and prediction accuracy and yield more interpretable results. However, these methods typically use graphs extracted from existing databases or rely on subject matter expertise, which are known to be incomplete and may contain false edges. To address this gap, we propose a graph-guided Bayesian modeling framework to account for network noise in regression models involving structured high-dimensional predictors. Specifically, we use 2 sources of network information, including the noisy graph extracted from existing databases and the estimated graph from observed predictors in the dataset at hand, to inform the model for the true underlying network via a latent scale modeling framework. This model is coupled with the Bayesian regression model with structured high-dimensional predictors involving an adaptive structured shrinkage prior. We develop an efficient Markov chain Monte Carlo algorithm for posterior sampling. We demonstrate the advantages of our method over existing methods in simulations, and through analyses of a genomics dataset and another proteomics dataset for Alzheimer's disease.

Modeling the risk of aquatic species invasion spread through boater movements and river connections.

Aquatic invasive species (AIS) are one of the greatest threats to the functioning of aquatic ecosystems worldwide. Once an invasive species has been introduced to a new region, many governments develop management strategies to reduce further spread. Nevertheless, managing AIS in a new region is challenging because of the vast areas that need protection and limited resources. Spatial heterogeneity in invasion risk is driven by environmental suitability and propagule pressure, which can be used to prioritize locations for surveillance and intervention activities. To better understand invasion risk across aquatic landscapes, we developed a simulation model to estimate the likelihood of a waterbody becoming invaded with an AIS. The model included waterbodies connected via a multilayer network that included boater movements and hydrological connections. In a case study of Minnesota, we used zebra mussels (Dreissena polymorpha) and starry stonewort (Nitellopsis obtusa) as model species. We simulated the impacts of management scenarios developed by stakeholders and created a decision-support tool available through an online application provided as part of the AIS Explorer dashboard. Our baseline model revealed that 89% of new zebra mussel invasions and 84% of new starry stonewort invasions occurred through boater movements, establishing it as a primary pathway of spread and offering insights beyond risk estimates generated by traditional environmental suitability models alone. Our results highlight the critical role of interventions applied to boater movements to reduce AIS dispersal.

Modelo del riesgo de la invasión de especies acuáticas dispersadas por movimiento de botes y conexiones entre ríos Resumen Las especies acuáticas invasoras (EAI) son una de las principales amenazas para el funcionamiento de los ecosistemas acuáticos a nivel mundial. Una vez que una especie invasora ha sido introducida a una nueva región, muchos gobiernos desarrollan estrategias de manejo para disminuir la dispersión. Sin embargo, el manejo de las especies acuáticas invasoras en una nueva región se complica debido a las amplias áreas que necesitan protección y los recursos limitados. La heterogeneidad espacial de un riesgo de invasión es causada por la idoneidad ambiental y la presión de propágulo, que puede usarse para priorizar la ubicación de las actividades de vigilancia e intervención. Desarrollamos una simulación para estimar la probabilidad de que un cuerpo de agua sea invadido por EAI para tener un mejor entendimiento del riesgo de invasión en los paisajes acuáticos. El modelo incluyó cuencas conectadas a través de una red multicapa que incluía movimiento de botes y conexiones hidrológicas. Usamos como especies modelo a Dreissena polymorpha y a Nitellopsis obtusa en un estudio de caso en Minnesota. Simulamos el impacto de los escenarios de manejo desarrollado por los actores y creamos una herramienta de decisiones por medio de una aplicación en línea proporcionada como parte del tablero del Explorer de EAI. Nuestro modelo de línea base reveló que el 89% de las invasiones nuevas de D. polymorpha y el 84% de las de N. obtusa ocurrieron debido al movimiento de los botes, lo que lo estableció como una vía primaria de dispersión y nos proporcionó información más allá de las estimaciones de riesgo generadas por los modelos tradicionales de idoneidad ambiental. Nuestros resultados resaltan el papel crítico de las intervenciones aplicadas al movimiento de los botes para reducir la dispersión de especies acuáticas invasoras.

Different computational relations in language are captured by distinct brain systems.

A critical way for humans to acquire information is through language, yet whether and how language experience drives specific neural semantic representations is still poorly understood. We considered statistical properties captured by 3 different computational principles of language (simple co-occurrence, network-(graph)-topological relations, and neural-network-vector-embedding relations) and tested the extent to which they can explain the neural patterns of semantic representations, measured by 2 functional magnetic resonance imaging experiments that shared common semantic processes. Distinct graph-topological word relations, and not simple co-occurrence or neural-network-vector-embedding relations, had unique explanatory power for the neural patterns in the anterior temporal lobe (capturing graph-common-neighbors), inferior frontal gyrus, and posterior middle/inferior temporal gyrus (capturing graph-shortest-path). These results were relatively specific to language: they were not explained by sensory-motor similarities and the same computational relations of visual objects (based on visual image database) showed effects in the visual cortex in the picture naming experiment. That is, different topological properties within language and the same topological computations (common-neighbors) for language and visual inputs are captured by different brain regions. These findings reveal the specific neural semantic representations along graph-topological properties of language, highlighting the information type-specific and statistical property-specific manner of semantic representations in the human brain.

White-matter degradation and dynamical compensation support age-related functional alterations in human brain.

Structural connectivity of the brain at different ages is analyzed using diffusion-weighted magnetic resonance imaging (MRI) data. The largest decrease of streamlines is found in frontal regions and for long inter-hemispheric links. The average length of the tracts also decreases, but the clustering is unaffected. From functional MRI we identify age-related changes of dynamic functional connectivity (dFC) and spatial covariation features of functional connectivity (FC) links captured by metaconnectivity. They indicate more stable dFC, but wider range and variance of MC, whereas static features of FC did not show any significant differences with age. We implement individual connectivity in whole-brain models and test several hypotheses for the mechanisms of operation among underlying neural system. We demonstrate that age-related functional fingerprints are only supported if the model accounts for: (i) compensation of the individual brains for the overall loss of structural connectivity and (ii) decrease of propagation velocity due to the loss of myelination. We also show that with these 2 conditions, it is sufficient to decompose the time-delays as bimodal distribution that only distinguishes between intra- and inter-hemispheric delays, and that the same working point also captures the static FC the best, and produces the largest variability at slow time-scales.

An auxiliary diagnostic tool for common fundus diseases based on fundus color photography and light-weight classification models.

OBJECTIVE: To evaluate the performance of two lightweight neural network models in the diagnosis of common fundus diseases and make comparison to another two classical models. METHODS: A total of 16,000 color fundus photography were collected, including 2000 each of glaucoma, diabetic retinopathy (DR), high myopia, central retinal vein occlusion (CRVO), age-related macular degeneration (AMD), optic neuropathy, and central serous chorioretinopathy (CSC), in addition to 2000 normal fundus. Fundus photography was obtained from patients or physical examiners who visited the Ophthalmology Department of Beijing Tongren Hospital, Capital Medical University. Each fundus photography has been diagnosed and labeled by two professional ophthalmologists. Two classical classification models (ResNet152 and DenseNet121), and two lightweight classification models (MobileNetV3 and ShufflenetV2), were trained. Area under the curve (AUC), sensitivity, specificity, accuracy, positive predictive value, and negative predictive value were used to evaluate the performance of the four models. RESULTS: Compared with the classical classification model, the total size and number of parameters of the two lightweight classification models were significantly reduced, and the classification speed was sharply improved. Compared with the DenseNet121 model, the ShufflenetV2 model took 50.7% less time to make a diagnosis on a fundus photography. The classical models performed better than lightweight classification models, and Densenet121 showed highest AUC in five out of the seven common fundus diseases. However, the performance of lightweight classification models is satisfying. The AUCs using MobileNetV3 model to diagnose AMD, diabetic retinopathy, glaucoma, CRVO, high myopia, optic atrophy, and CSC were 0.805, 0.892, 0.866, 0.812, 0.887, 0.868, and 0.803, respectively. For ShufflenetV2model, the AUCs for the above seven diseases were 0.856, 0.893, 0.855, 0.884, 0.891, 0.867, and 0.844, respectively. CONCLUSION: The training of light-weight neural network models based on color fundus photography for the diagnosis of common fundus diseases is not only fast but also has a significant reduction in storage size and parameter number compared with the classical classification model, and can achieve satisfactory accuracy.

Adaptive staffing can mitigate essential worker disease and absenteeism in an emerging epidemic.

Essential worker absenteeism has been a pressing problem in the COVID-19 pandemic. Nearly 20% of US hospitals experienced staff shortages, exhausting replacement pools and at times requiring COVID-positive healthcare workers to remain at work. To our knowledge there are no data-informed models examining how different staffing strategies affect epidemic dynamics on a network in the context of rising worker absenteeism. Here we develop a susceptible-infected-quarantined-recovered adaptive network model using pair approximations to gauge the effects of worker replacement versus redistribution of work among remaining healthy workers in the early epidemic phase. Parameterized with hospital data, the model exhibits a time-varying trade-off: Worker replacement minimizes peak prevalence in the early phase, while redistribution minimizes final outbreak size. Any "ideal" strategy requires balancing the need to maintain a baseline number of workers against the desire to decrease total number infected. We show that one adaptive strategy-switching from replacement to redistribution at epidemic peak-decreases disease burden by 9.7% and nearly doubles the final fraction of healthy workers compared to pure replacement.