Learning to Learn Task-Adaptive Hyperparameters for Few-Shot Learning.

The objective of few-shot learning is to design a system that can adapt to a given task with only few examples while achieving generalization. Model-agnostic meta-learning (MAML), which has recently gained the popularity for its simplicity and flexibility, learns a good initialization for fast adaptation to a task under few-data regime. However, its performance has been relatively limited especially when novel tasks are different from tasks previously seen during training. In this work, instead of searching for a better initialization, we focus on designing a better fast adaptation process. Consequently, we propose a new task-adaptive weight update rule that greatly enhances the fast adaptation process. Specifically, we introduce a small meta-network that can generate per-step hyperparameters for each given task: learning rate and weight decay coefficients. The experimental results validate that learning a good weight update rule for fast adaptation is the equally important component that has drawn relatively less attention in the recent few-shot learning approaches. Surprisingly, fast adaptation from random initialization with ALFA can already outperform MAML. Furthermore, the proposed weight-update rule is shown to consistently improve the task-adaptation capability of MAML across diverse problem domains: few-shot classification, cross-domain few-shot classification, regression, visual tracking, and video frame interpolation.

Evaluating Z-FA-FMK, a host cathepsin L protease inhibitor, as a potent and broad-spectrum antiviral therapy against SARS-CoV-2 and related coronaviruses.

Even though the World Health Organization announced the end of the COVID-19 pandemic as a global public health emergency on May 5, 2023, SARS-CoV-2 continues to pose a significant health threat worldwide, resulting in substantial numbers of infections and fatalities. This study investigated the antiviral potential of Z-FA-FMK (FMK), a novel host cathepsin L protease inhibitor, against SARS-CoV-2 infection using both in vitro and in vivo models. In vitro assessments of FMK against a diverse set of SARS-CoV-2 strains, including the Wuhan-like strain and nine variants, demonstrated potent inhibition with EC50 values ranging from 0.55 to 2.41 µM, showcasing similar or superior efficacy compared to FDA-approved antivirals nirmatrelvir (NTV) and molnupiravir (MPV). In vivo experiments using orally administered FMK (25 mg/kg) in SARS-CoV-2-infected K18 hACE2 transgenic mice revealed improved survival rates of 60% and accelerated recovery compared to NTV and MPV treatments. Additionally, FMK displayed a longer half-life (17.26 ± 8.89 h) than NTV and MPV in the mouse model. Due to its host-targeting mechanism, FMK offers potential advantages such as reduced drug resistance and broad-spectrum antiviral activity against multiple coronaviruses. These findings indicate that FMK may serve as a promising candidate for further clinical evaluation in the fight against SARS-CoV-2.

Reciprocal enhancement of SARS-CoV-2 and influenza virus replication in human pluripotent stem cell-derived lung organoids1.

Patients with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and influenza A virus (FLUAV) coinfections were associated with severe respiratory failure and more deaths. Here, we developed a model for studying SARS-CoV-2 and FLUAV coinfection using human pluripotent stem cell-induced alveolar type II organoids (hiAT2). hiAT2 organoids were susceptible to infection by both viruses and had features of severe lung damage. A single virus markedly enhanced the susceptibility to other virus infections. SARS-CoV-2 delta variants upregulated α-2-3-linked sialic acid, while FLUAV upregulated angiotensin-converting enzyme 2 (ACE2) and transmembrane serine protease 2 (TMPRSS2). Moreover, coinfection by SARS-CoV-2 and FLUAV caused hyperactivation of proinflammatory and immune-related signaling pathways and cellular damage compared to a respective single virus in hiAT2 organoids. This study provides insight into molecular mechanisms underlying enhanced infectivity and severity in patients with co-infection of SARS-CoV-2 and FLUAV, which may aid in the development of therapeutics for such co-infection cases.

Global Context Attention for Robust Visual Tracking.

Although there have been recent advances in Siamese-network-based visual tracking methods where they show high performance metrics on numerous large-scale visual tracking benchmarks, persistent challenges regarding the distractor objects with similar appearances to the target object still remain. To address these aforementioned issues, we propose a novel global context attention module for visual tracking, where the proposed module can extract and summarize the holistic global scene information to modulate the target embedding for improved discriminability and robustness. Our global context attention module receives a global feature correlation map to elicit the contextual information from a given scene and generates the channel and spatial attention weights to modulate the target embedding to focus on the relevant feature channels and spatial parts of the target object. Our proposed tracking algorithm is tested on large-scale visual tracking datasets, where we show improved performance compared to the baseline tracking algorithm while achieving competitive performance with real-time speed. Additional ablation experiments also validate the effectiveness of the proposed module, where our tracking algorithm shows improvements in various challenging attributes of visual tracking.

Single Camera-Based Dual-Channel Near-Infrared Fluorescence Imaging system.

In this study, we propose a single camera-based dual-channel near-infrared (NIR) fluorescence imaging system that produces color and dual-channel NIR fluorescence images in real time. To simultaneously acquire color and dual-channel NIR fluorescence images of two fluorescent agents, three cameras and additional optical parts are generally used. As a result, the volume of the image acquisition unit increases, interfering with movements during surgical procedures and increasing production costs. In the system herein proposed, instead of using three cameras, we set a single camera equipped with two image sensors that can simultaneously acquire color and single-channel NIR fluorescence images, thus reducing the volume of the image acquisition unit. The single-channel NIR fluorescence images were time-divided into two channels by synchronizing the camera and two excitation lasers, and the noise caused by the crosstalk effect between the two fluorescent agents was removed through image processing. To evaluate the performance of the system, experiments were conducted for the two fluorescent agents to measure the sensitivity, crosstalk effect, and signal-to-background ratio. The compactness of the resulting image acquisition unit alleviates the inconvenient movement obstruction of previous devices during clinical and animal surgery and reduces the complexity and costs of the manufacturing process, which may facilitate the dissemination of this type of system.

Combination therapy with nirmatrelvir and molnupiravir improves the survival of SARS-CoV-2 infected mice.

As the SARS-CoV-2 pandemic remains uncontrolled owing to the continuous emergence of variants of concern, there is an immediate need to implement the most effective antiviral treatment strategies, especially for risk groups. Here, we evaluated the therapeutic potency of nirmatrelvir, remdesivir and molnupiravir, and their combinations in SARS-CoV-2 infected K18-hACE2 transgenic mice. Systemic treatment of mice with each drug (20 mg/kg) resulted in slightly enhanced antiviral efficacy and yielded an increased life expectancy of only about 20-40% survival. However, combination therapy with nirmatrelvir (20 mg/kg) and molnupiravir (20 mg/kg) in lethally infected mice showed profound inhibition of SARS-CoV-2 replication in both the lung and brain and synergistically improved survival rates up to 80% compared to those with nirmatrelvir (36%, P < 0.001) and molnupiravir (43%, P < 0.001) administered alone. This combination therapy effectively reduced clinical severity score, virus-induced tissue damage, and viral distribution compared to those in animals treated with these monotherapies. Furthermore, all these assessments associated with this combination were also significantly higher than that of mice receiving remdesivir monotherapy (P < 0.001) and the nirmatrelvir (20 mg/kg) and remdesivir (20 mg/kg) combination (P < 0.001), underscored the clinical significance of this combination. By contrast, the nirmatrelvir and remdesivir combination showed less antiviral efficacy, with lower survival compared to nirmatrelvir monotherapy due to the insufficient plasma exposure of the remdesivir, demonstrating the inefficient therapeutic effect of this combination in the mouse model. The combination therapy with nirmatrelvir and molnupiravir contributes to alleviated morbidity and mortality, which can serve as a basis for the design of clinical studies of this combination in the treatment of COVID-19 patients.

SARS-CoV-2 spike trimer vaccine expressed in Nicotiana benthamiana adjuvanted with Alum elicits protective immune responses in mice.

The ongoing coronavirus disease 2019 (COVID-19) pandemic has spurred rapid development of vaccines as part of the public health response. However, the general strategy used to construct recombinant trimeric severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) spike (S) proteins in mammalian cells is not completely adaptive to molecular farming. Therefore, we generated several constructs of recombinant S proteins for high expression in Nicotiana benthamiana. Intramuscular injection of N. benthamiana-expressed Sct vaccine (NSct Vac) into Balb/c mice elicited both humoral and cellular immune responses, and booster doses increased neutralizing antibody titres. In human angiotensin-converting enzyme knock-in mice, two doses of NSct Vac induced anti-S and neutralizing antibodies, which cross-neutralized Alpha, Beta, Delta and Omicron variants. Survival rates after lethal challenge with SARS-CoV-2 were up to 80%, without significant body weight loss, and viral titres in lung tissue fell rapidly, with no infectious virus detectable at 7-day post-infection. Thus, plant-derived NSct Vac could be a candidate COVID-19 vaccine.

Development of Intraoperative Near-Infrared Fluorescence Imaging System Using a Dual-CMOS Single Camera.

We developed a single-camera-based near-infrared (NIR) fluorescence imaging device using indocyanine green (ICG) NIR fluorescence contrast agents for image-induced surgery. In general, a fluorescent imaging system that simultaneously provides color and NIR images uses two cameras, which is disadvantageous because it increases the imaging head of the system. Recently, a single-camera-based NIR optical imaging device with quantum efficiency partially extended to the NIR region was developed to overcome this drawback. The system used RGB_NIR filters for camera sensors to provide color and NIR images simultaneously; however, the sensitivity and resolution of the infrared images are reduced by 1/4, and the exposure time and gain cannot be set individually when acquiring color and NIR images. Thus, to overcome these shortcomings, this study developed a compact fluorescent imaging system that uses a single camera with two complementary metal-oxide semiconductor (CMOS) image sensors. Sensitivity and signal-to-background ratio were measured according to the concentrations of ICG solution, exposure time, and camera gain to evaluate the performance of the imaging system. Consequently, the clinical applicability of the system was confirmed through the toxicity analysis of the light source and in vivo testing.

Test-Time Adaptation for Video Frame Interpolation via Meta-Learning.

Video frame interpolation is a challenging problem that involves various scenarios depending on the variety of foreground and background motions, frame rate, and occlusion. Therefore, generalizing across different scenes is difficult for a single network with fixed parameters. Ideally, one could have a different network for each scenario, but this will be computationally infeasible for practical applications. In this work, we propose MetaVFI, an adaptive video frame interpolation algorithm that uses additional information readily available at test time but has not been exploited in previous works. We initially show the benefits of test-time adaptation through simple fine-tuning of a network and then greatly improve its efficiency by incorporating meta-learning. Thus, we obtain significant performance gains with only a single gradient update without introducing any additional parameters. Moreover, the proposed MetaVFI algorithm is model-agnostic which can be easily combined with any video frame interpolation network. We show that our adaptive framework greatly improves the performance of baseline video frame interpolation networks on multiple benchmark datasets.

An infectious cDNA clone of a growth attenuated Korean isolate of MERS coronavirus KNIH002 in clade B.

The MERS-CoV isolated during the 2015 nosocomial outbreak in Korea showed distinctive differences in mortality and transmission patterns compared to the prototype MERS-CoV EMC strain belonging to clade A. We established a BAC-based reverse genetics system for a Korean isolate of MERS-CoV KNIH002 in the clade B phylogenetically far from the EMC strain, and generated a recombinant MERS-CoV expressing red fluorescent protein. The virus rescued from the infectious clone and KNIH002 strain displayed growth attenuation compared to the EMC strain. Consecutive passages of the rescued virus rapidly generated various ORF5 variants, highlighting its genetic instability and calling for caution in the use of repeatedly passaged virus in pathogenesis studies and for evaluation of control measures against MERS-CoV. The infectious clone for the KNIH002 in contemporary epidemic clade B would be useful for better understanding of a functional link between molecular evolution and pathophysiology of MERS-CoV by comparative studies with EMC strain.

Antiviral activity of digoxin and ouabain against SARS-CoV-2 infection and its implication for COVID-19.

The current coronavirus (COVID-19) pandemic is exacerbated by the absence of effective therapeutic agents. Notably, patients with COVID-19 and comorbidities such as hypertension and cardiac diseases have a higher mortality rate. An efficient strategy in response to this issue is repurposing drugs with antiviral activity for therapeutic effect. Digoxin (DIG) and ouabain (OUA) are FDA drugs for heart diseases that have antiviral activity against several coronaviruses. Thus, we aimed to assess antiviral activity of DIG and OUA against SARS-CoV-2 infection. The half-maximal inhibitory concentrations (IC50) of DIG and OUA were determined at a nanomolar concentration. Progeny virus titers of single-dose treatment of DIG, OUA and remdesivir were approximately 103-, 104- and 103-fold lower (> 99% inhibition), respectively, than that of non-treated control or chloroquine at 48 h post-infection (hpi). Furthermore, therapeutic treatment with DIG and OUA inhibited over 99% of SARS-CoV-2 replication, leading to viral inhibition at the post entry stage of the viral life cycle. Collectively, these results suggest that DIG and OUA may be an alternative treatment for COVID-19, with potential additional therapeutic effects for patients with cardiovascular disease.

Simple Shading Correction Method for Brightfield Whole Slide Imaging.

Whole slide imaging (WSI) refers to the process of creating a high-resolution digital image of a whole slide. Since digital images are typically produced by stitching image sequences acquired from different fields of view, the visual quality of the images can be degraded owing to shading distortion, which produces black plaid patterns on the images. A shading correction method for brightfield WSI is presented, which is simple but robust not only against typical image artifacts caused by specks of dust and bubbles, but also against fixed-pattern noise, or spatial variations in pixel values under uniform illumination. The proposed method comprises primarily of two steps. The first step constructs candidates of a shading distortion model from a stack of input image sequences. The second step selects the optimal model from the candidates. The proposed method was compared experimentally with two previous state-of-the-art methods, regularized energy minimization (CIDRE) and background and shading correction (BaSiC) and showed better correction scores, as smooth operations and constraints were not imposed when estimating the shading distortion. The correction scores, averaged over 40 image collections, were as follows: proposed method, 0.39 ± 0.099; CIDRE method, 0.67 ± 0.047; BaSiC method, 0.55 ± 0.038. Based on the quantitative evaluations, we can confirm that the proposed method can correct not only shading distortion, but also fixed-pattern noise, compared with the two previous state-of-the-art methods.

Characterization of a human monoclonal antibody generated from a B-cell specific for a prefusion-stabilized spike protein of Middle East respiratory syndrome coronavirus.

Middle East respiratory syndrome coronavirus (MERS-CoV) causes severe respiratory infection and continues to infect humans, thereby contributing to a high mortality rate (34.3% in 2019). In the absence of an available licensed vaccine and antiviral agent, therapeutic human antibodies have been suggested as candidates for treatment. In this study, human monoclonal antibodies were isolated by sorting B cells from patient's PBMC cells with prefusion stabilized spike (S) probes and a direct immunoglobulin cloning strategy. We identified six receptor-binding domain (RBD)-specific and five S1 (non-RBD)-specific antibodies, among which, only the RBD-specific antibodies showed high neutralizing potency (IC50 0.006-1.787 µg/ml) as well as high affinity to RBD. Notably, passive immunization using a highly potent antibody (KNIH90-F1) at a relatively low dose (2 mg/kg) completely protected transgenic mice expressing human DPP4 against MERS-CoV lethal challenge. These results suggested that human monoclonal antibodies isolated by using the rationally designed prefusion MERS-CoV S probe could be considered potential candidates for the development of therapeutic and/or prophylactic antiviral agents for MERS-CoV human infection.

The effect of mutations derived from mouse-adapted H3N2 seasonal influenza A virus to pathogenicity and host adaptation.

Elucidating the genetic basis of influenza A viruses (IAVs) is important to understand which mutations will determine the virulence and the host range of mammals. Here, seasonal H3N2 influenza was adapted in mice by serial passage and four mutants, each carrying amino acid substitutions related to mouse adaptation in either the PB2, HA, NP, or NA protein, were generated. To confirm the contribution of each gene to enhanced pathogenicity and mouse adaptation, mice were inoculated with the respective variants, and virulence, replication, histopathology, and infectivity were examined. The virus harboring HA mutations displayed increased infection efficiency and replication competence, resulting in higher mortality in mice relative to those infected with wild-type virus. By contrast, the NP D34N mutation caused rapid and widespread infection in multiple organs without presenting virulent symptoms. Additionally, the PB2 F323L mutation presented delayed but elevated replication competence in the respiratory tract, whereas the S331R mutation in NA showed no considerable effects on mouse adaptation. These results suggested that mouse-adapted changes in HA are major factors in increased pathogenicity and that mutations in NP and PB2 also contribute to cross-species adaptability. Our findings offer a better understanding of the molecular basis for IAV pathogenicity and adaptation in a new host.

Drug Discovery Platform Targeting M. tuberculosis with Human Embryonic Stem Cell-Derived Macrophages.

A major limitation in anti-tuberculosis drug screening is the lack of reliable and scalable models for homogeneous human primary macrophage cells of non-cancer origin. Here we report a modified protocol for generating homogeneous populations of macrophage-like cells from human embryonic stem cells. The induced macrophages, referred to as iMACs, presented similar transcriptomic profiles and characteristic immunological features of classical macrophages and were permissive to viral and bacterial infection, in particular Mycobacterium tuberculosis (Mtb). More importantly, iMAC production was amenable to scale up. To evaluate iMAC efficiency in high-throughput anti-tuberculosis drug screening, we performed a phenotypic screening against intracellular Mtb, involving a library of 3,716 compounds that included FDA-approved drugs and other bioactive compounds. Our primary screen identified 120 hits, which were validated in a secondary screen by dose-intracellular and -extracellular Mtb assays. Our confirmatory studies identified a novel anti-Mtb compound, 10-DEBC, also showing activity against drug-resistant strains.

Nation-wide surveillance of human acute respiratory virus infections between 2013 and 2015 in Korea.

The prevalence of eight respiratory viruses detected in patients with acute respiratory infections (ARIs) in Korea was investigated through analysis of data recorded by the Korea Influenza and Respiratory Viruses Surveillance System (KINRESS) from 2013 to 2015. Nasal aspirate and throat swabs specimens were collected from 36 915 patients with ARIs, and viral nucleic acids were detected by real-time (reverse-transcription) polymerase chain reaction for eight respiratory viruses, including human respiratory syncytial viruses (HRSVs), influenza viruses (IFVs), human parainfluenza viruses (HPIVs), human coronaviruses (HCoVs), human rhinovirus (HRV), human adenovirus (HAdV), human bocavirus (HBoV), and human metapneumovirus (HMPV). The overall positive rate of patient specimens was 49.4% (18 236/36 915), 5% of which carried two or more viruses simultaneously. HRV (15.6%) was the most predominantly detected virus, followed by IFVs (14.6%), HAdV (7.5%), HPIVs (5.8%), HCoVs (4.2%), HRSVs (3.6%), HBoV (1.9%), and HMPV (1.6%). Most of the ARIs were significantly correlated with clinical symptoms of fever, cough, and runny nose. Although HRV and HAdV were frequently detected throughout the year in patients, other respiratory viruses showed apparent seasonality. HRSVs and IFVs were the major causative agents of acute respiratory diseases in infants and young children. Overall, this study demonstrates a meaningful relationship between viral infection and typical manifestations of known clinical features as well as seasonality, age distribution, and co-infection among respiratory viruses. Therefore, these data could provide useful information for public health management and to enhance patient care for primary clinicians.

Oseltamivir-resistant influenza viruses isolated in South Korea from 2005 to 2010.

South Korean isolates of oseltamivir-resistant influenza viruses from 2005-2010 were investigated with a total 491 influenza viruses identified from 1702 specimens. Neuraminidase genes from 342 influenza viruses (71 A/H1N1, 74 pandemic A/H1N1 2009, 117 A/H3N2, and 80 B) were analyzed by RT-PCR with molecular markers for oseltamivir resistance. The H274Y mutation in the NA protein was identified in 100 % (n=40) of A/H1N1 viruses circulating in 2008-2009. Influenza A/H1N1 viruses harboring the H274Y substitution exhibited, on average, a 626-fold reduction in oseltamivir susceptibility and clustered with the A/Norway/1736/2007 strain. Close and timely monitoring for resistance to clinically available influenza antivirals should be consistently performed.

Development and evaluation of multiplex real-time RT-PCR assays for seasonal, pandemic A/H1pdm09 and avian A/H5 influenza viruses detection.

Since the pandemic influenza A (H1N1) 2009 ((H1N1)pdm09) virus spread all over the world, the (H1N1)pdm09 virus has been circulating with seasonal influenza viruses. We developed rapid and sensitive one-step multiplex real-time RT-PCR assays (rRT-PCR) for simultaneous detection of influenza viruses currently circulating in humans, and the avian A/H5 virus. The detection limit of each assay was 4.8 to 1 copies per reaction and no cross-reactivity with other major respiratory pathogens was found. Analytical positive predictive value (PPV), negative predictive value (NPV) sensitivity and specificity were 100%, 94.1%, 93.7% and 100%, respectively. Clinical evaluation revealed that 1,976 (16.5%) of 11,963 throat swabs from patients with respiratory symptoms were confirmed as 1,651 (83.6%) A/H1pdm09, 308 (15.6%) A/H3 and 17 (0.8%) B virus during the 2010-2011 influenza season. Collectively, the multiplex rRT-PCR assays described here provide a practical tool for reliable implementation of influenza surveillance and diagnosis.

Impact of influenza virus escape-mutations on influenza detection by the rapid influenza diagnostic test.

During the influenza pandemic of 2009-2010, rapid influenza diagnostic tests (RIDTs) were used to detect influenza viral infections because they are quick and simple to use. However, retrospective studies showed that RIDTs performed poorly when used to diagnose pandemic viral infections. Determining how amino acid sequence changes in pandemic or epidemic influenza viral antigens impact clinical value of RIDTs has not been possible, because the viral epitopes recognized by RIDTs have been not mapped. In this study, the effect of escape-variations or mutations in influenza viral antigens upon the sensitivity and specificity of an RIDT was investigated by characterizing the immunological properties of the antibodies used in the RIDT. Escape-mutants were generated by cultivating A/Korea/01/2009 in the presence of an excess of the same antibodies used in the RIDT. Escape-mutants not recognized by the RIDT were selected. Epitopes recognized by the RIDT were mapped by comparing the sequence and immunological analysis of the escape-variants and wild-type isolates. The RIDT antibodies recognized epitopes on the Sa antigenic site and in the F subdomain in hemagglutinin. Variants bearing mutations in these epitopes were not detected by the RIDT. The frequency of escape-variants emerging since the 2009-2010 pandemic was calculated as 1.27% using in silico surveillance of influenza sequence databases. These results suggest that mapping the relevant epitopes of RIDTs and making such information available to clinics would be helpful for determining whether RIDTs match newly emergent strains and subtypes prior to retrospective re-evaluation of the RIDTs using clinical specimens.

High prevalence of amantadine-resistant influenza A virus isolated in Gyeonggi Province, South Korea, during 2005-2010.

Amantadine resistance among influenza A viruses was investigated in South Korea in 2005-2010. Of 308 influenza A viruses examined, 229 had the S31N substitution in the M2 protein. The frequency of amantadine resistance was 30 %, 100 %, and 76 % in influenza A/H1N1, pandemic A/H1N1 2009(A/H1N1pdm), and A/H3N2 subtypes, respectively. The amantadine-resistant influenza A/H1N1pdm and A/H3N2 viruses were circulating continuously from 2008 to 2009 and from 2005 to 2006, respectively. Amantadine resistance among influenza A viruses increased dramatically during the 5-year study period, and this has diminished the usefulness of this class of drugs.