Mapping of H3N2 influenza antigenic evolution in China reveals a strategy for vaccine strain recommendation

Global antigenic landscape and vaccine recommendation strategy for low pathogenic avian influenza A (H9N2) viruses

The sustained circulation of H9N2 avian influenza viruses (AIVs) poses a significant threat for contributing to a new pandemic. Given the temporal and spatial uncertainty in the antigenicity of H9N2 AIVs, the immune protection efficiency of vaccines remains challenging. By developing an antigenicity prediction method for H9N2 AIVs, named PREDAC-H9, the global antigenic landscape of H9N2 AIVs was mapped. PREDAC-H9 utilizes the XGBoost model with 14 well-designed features. The XGBoost model was built and evaluated to predict the antigenic relationship between any two viruses with high values of 81.1 %, 81.4 %, 81.3 %, 81.1 %, and 89.4 % in accuracy, precision, recall, F1 value, and area under curve (AUC), respectively. Then the antigenic correlation network (ACnet) was constructed based on the predicted antigenic relationship for H9N2 AIVs from 1966 to 2022, and ten major antigenic clusters were identified. Of these, four novel clusters were generated in China in the past decade, demonstrating the unique complex situation there. To help tackle this situation, we applied PREDAC-H9 to calculate the cluster-transition determining sites and screen out virus strains with the high cross-protective spectrum, thus providing an in silico reference for vaccine recommendation. The proposed model will reduce the clinical monitoring workload and provide a useful tool for surveillance and control of H9N2 AIVs.

AAontology: An ontology of amino acid scales for interpretable machine learning

Amino acid scales are crucial for protein prediction tasks, many of them being curated in the AAindex database. Despite various clustering attempts to organize them and to better understand their relationships, these approaches lack the fine-grained classification necessary for satisfactory interpretability in many protein prediction problems. To address this issue, we developed AAontology-a two-level classification for 586 amino acid scales (mainly from AAindex) together with an in-depth analysis of their relations-using bag-of-word-based classification, clustering, and manual refinement over multiple iterations. AAontology organizes physicochemical scales into 8 categories and 67 subcategories, enhancing the interpretability of scale-based machine learning methods in protein bioinformatics. Thereby it enables researchers to gain a deeper biological insight. We anticipate that AAontology will be a building block to link amino acid properties with protein function and dysfunctions as well as aid informed decision-making in mutation analysis or protein drug design.

Seasonal antigenic prediction of influenza A H3N2 using machine learning

Antigenic characterization of circulating influenza A virus (IAV) isolates is routinely assessed by using the hemagglutination inhibition (HI) assays for surveillance purposes. It is also used to determine the need for annual influenza vaccine updates as well as for pandemic preparedness. Performing antigenic characterization of IAV on a global scale is confronted with high costs, animal availability, and other practical challenges. Here we present a machine learning model that accurately predicts (normalized) outputs of HI assays involving circulating human IAV H3N2 viruses, using their hemagglutinin subunit 1 (HA1) sequences and associated metadata. Each season, the model learns an updated nonlinear mapping of genetic to antigenic changes using data from past seasons only. The model accurately distinguishes antigenic variants from non-variants and adaptively characterizes seasonal dynamics of HA1 sites having the strongest influence on antigenic change. Antigenic predictions produced by the model can aid influenza surveillance, public health management, and vaccine strain selection activities.

Reconstructed influenza A/H3N2 infection histories reveal variation in incidence and antibody dynamics over the life course

Humans experience many influenza infections over their lives, resulting in complex and varied immunological histories. Although experimental and quantitative analyses have improved our understanding of the immunological processes defining an individual's antibody repertoire, how these within-host processes are linked to population-level influenza epidemiology remains unclear. Here, we used a multi-level mathematical model to jointly infer antibody dynamics and individual-level lifetime influenza A/H3N2 infection histories for 1,130 individuals in Guangzhou, China, using 67,683 haemagglutination inhibition (HI) assay measurements against 20 A/H3N2 strains from repeat serum samples collected between 2009 and 2015. These estimated infection histories allowed us to reconstruct historical seasonal influenza patterns and to investigate how influenza incidence varies over time, space and age in this population. We estimated median annual influenza infection rates to be approximately 18% from 1968 to 2015, but with substantial variation between years. 88% of individuals were estimated to have been infected at least once during the study period (2009-2015), and 20% were estimated to have three or more infections in that time. We inferred decreasing infection rates with increasing age, and found that annual attack rates were highly correlated across all locations, regardless of their distance, suggesting that age has a stronger impact than fine-scale spatial effects in determining an individual's antibody profile. Finally, we reconstructed each individual's expected antibody profile over their lifetime and inferred an age-stratified relationship between probability of infection and HI titre. Our analyses show how multi-strain serological panels provide rich information on long term, epidemiological trends, within-host processes and immunity when analyzed using appropriate inference methods, and adds to our understanding of the life course epidemiology of influenza A/H3N2.

Spatiotemporal genotype replacement of H5N8 avian influenza viruses contributed to H5N1 emergence in 2021/2022 panzootic

Since 2020, clade 2.3.4.4b highly pathogenic avian influenza H5N8 and H5N1 viruses have swept through continents, posing serious threats to the world. Through comprehensive analyses of epidemiological, genetic, and bird migration data, we found that the dominant genotype replacement of the H5N8 viruses in 2020 contributed to the H5N1 outbreak in the 2021/2022 wave. The 2020 outbreak of the H5N8 G1 genotype instead of the G0 genotype produced reassortment opportunities and led to the emergence of a new H5N1 virus with G1's HA and MP genes. Despite extensive reassortments in the 2021/2022 wave, the H5N1 virus retained the HA and MP genes, causing a significant outbreak in Europe and North America. Furtherly, through the wild bird migration flyways investigation, we found that the temporal-spatial coincidence between the outbreak of the H5N8 G1 virus and the bird autumn migration may have expanded the H5 viral spread, which may be one of the main drivers of the emergence of the 2020-2022 H5 panzootic.IMPORTANCESince 2020, highly pathogenic avian influenza (HPAI) H5 subtype variants of clade 2.3.4.4b have spread across continents, posing unprecedented threats globally. However, the factors promoting the genesis and spread of H5 HPAI viruses remain unclear. Here, we found that the spatiotemporal genotype replacement of H5N8 HPAI viruses contributed to the emergence of the H5N1 variant that caused the 2021/2022 panzootic, and the viral evolution in poultry of Egypt and surrounding area and autumn bird migration from the Russia-Kazakhstan region to Europe are important drivers of the emergence of the 2020-2022 H5 panzootic. These findings provide important targets for early warning and could help control the current and future HPAI epidemics.

MAIVeSS: streamlined selection of antigenically matched, high-yield viruses for seasonal influenza vaccine production

Vaccines are the main pharmaceutical intervention used against the global public health threat posed by influenza viruses. Timely selection of optimal seed viruses with matched antigenicity between vaccine antigen and circulating viruses and with high yield underscore vaccine efficacy and supply, respectively. Current methods for selecting influenza seed vaccines are labor intensive and time-consuming. Here, we report the Machine-learning Assisted Influenza VaccinE Strain Selection framework, MAIVeSS, that enables streamlined selection of naturally circulating, antigenically matched, and high-yield influenza vaccine strains directly from clinical samples by using molecular signatures of antigenicity and yield to support optimal candidate vaccine virus selection. We apply our framework on publicly available sequences to select A(H1N1)pdm09 vaccine candidates and experimentally confirm that these candidates have optimal antigenicity and growth in cells and eggs. Our framework can potentially reduce the optimal vaccine candidate selection time from months to days and thus facilitate timely supply of seasonal vaccines.

A sequence-based machine learning model for predicting antigenic distance for H3N2 influenza virus

Introduction

Seasonal influenza A H3N2 viruses are constantly changing, reducing the effectiveness of existing vaccines. As a result, the World Health Organization (WHO) needs to frequently update the vaccine strains to match the antigenicity of emerged H3N2 variants. Traditional assessments of antigenicity rely on serological methods, which are both labor-intensive and time-consuming. Although numerous computational models aim to simplify antigenicity determination, they either lack a robust quantitative linkage between antigenicity and viral sequences or focus restrictively on selected features.

Methods

Here, we propose a novel computational method to predict antigenic distances using multiple features, including not only viral sequence attributes but also integrating four distinct categories of features that significantly affect viral antigenicity in sequences.

Results

This method exhibits low error in virus antigenicity prediction and achieves superior accuracy in discerning antigenic drift. Utilizing this method, we investigated the evolution process of the H3N2 influenza viruses and identified a total of 21 major antigenic clusters from 1968 to 2022.

Discussion

Interestingly, our predicted antigenic map aligns closely with the antigenic map generated with serological data. Thus, our method is a promising tool for detecting antigenic variants and guiding the selection of vaccine candidates.

Inhibition of miR-200b-3p confers broad-spectrum resistance to viral infection by targeting TBK1

The host innate immune system's defense against viral infections depends heavily on type I interferon (IFN-I) production. Research into the mechanisms of virus-host interactions is essential for developing novel antiviral therapies. In this study, we compared the effect of the five members of the microRNA-200 (miR-200) family on IFN-I production during viral infection and found that miR-200b-3p displayed the most pronounced regulatory effect. During viral infection, we discovered that the transcriptional level of microRNA-200b-3p (miR-200b-3p) increased with the infection of influenza virus (IAV) and vesicular stomatitis virus (VSV), and miR-200b-3p production was modulated by the activation of the ERK and p38 pathways. We identified cAMP response element binding protein (CREB) as a novel transcription factor that binds to the miR-200b-3p promoter. MiR-200b-3p reduces NF-κB and IRF3-mediated IFN-I production by targeting the 3' untranslated region (3' UTR) of TBK1 mRNA. Applying miR-200b-3p inhibitor enhances IFN-I production in IAV and VSV-infected mouse models, thus inhibiting viral replication and improving mouse survival ratio. Importantly, in addition to IAV and VSV, miR-200b-3p inhibitors exhibited potent antiviral effects against multiple pathogenic viruses threatening human health worldwide. Overall, our study suggests that miR-200b-3p might be a potential therapeutic target for broad-spectrum antiviral therapy. IMPORTANCE The innate immune response mediated by type I interferon (IFN-I) is essential for controlling viral replication. MicroRNAs (miRNAs) have been found to regulate the IFN signaling pathway. In this study, we describe a novel function of miRNA-200b-3p in negatively regulating IFN-I production during viral infection. miRNA-200b-3p was upregulated by the MAPK pathway activated by IAV and VSV infection. The binding of miRNA-200b-3p to the 3' UTR of TBK1 mRNA reduced IFN-I activation mediated by IRF3 and NF-κB. Application of miR-200b-3p inhibitors exhibited potent antiviral effects against multiple RNA and DNA viruses. These results provide fresh insight into understanding the impact of miRNAs on host-virus interactions and reveal a potential therapeutic target for common antiviral intervention.

Development of PREDAC-H1pdm to model the antigenic evolution of influenza A/(H1N1) pdm09 viruses

The Influenza A (H1N1) pdm09 virus caused a global pandemic in 2009 and has circulated seasonally ever since. As the continual genetic evolution of hemagglutinin in this virus leads to antigenic drift, rapid identification of antigenic variants and characterization of the antigenic evolution are needed. In this study, we developed PREDAC-H1pdm, a model to predict antigenic relationships between H1N1pdm viruses and identify antigenic clusters for post-2009 pandemic H1N1 strains. Our model performed well in predicting antigenic variants, which was helpful in influenza surveillance. By mapping the antigenic clusters for H1N1pdm, we found that substitutions on the Sa epitope were common for H1N1pdm, whereas for the former seasonal H1N1, substitutions on the Sb epitope were more common in antigenic evolution. Additionally, the localized epidemic pattern of H1N1pdm was more obvious than that of the former seasonal H1N1, which could make vaccine recommendation more sophisticated. Overall, the antigenic relationship prediction model we developed provides a rapid determination method for identifying antigenic variants, and the further analysis of evolutionary and epidemic characteristics can facilitate vaccine recommendations and influenza surveillance for H1N1pdm.

Effect of difference between EV-A71 virus epidemic strain and "vaccine strain" on neutralizing antibody titer

Hand, foot and mouth disease was mainly caused by EV-A71 virus. The main antigen structure of VP1 region of EV-A71 was easily varied. Here, we investigated the seroprevalence of EV-A71 based on a large group of healthy individuals in Beijing, China, in order to study the effectiveness of EV-A71 vaccine in a real-world setting. BrCr and the clinical strain isolated from the Chinese mainland in 2008 ("vaccine strain:"CMU4232/BJ/CHN/2008), EV-A71 C4 epidemic strains isolated in 2010, 2013, and 2016, were tested for neutralizing antibodies (NtAb) in every year. Phylogenetic tree analysis of the EV-A71 strains above, as well as amino acid composition homologous sequence analysis were applied. The "vaccine strain" has 83.0% homology with FY23, H07 and FY7VP5. It belongs to the same branch of C4a as 10 C4, 13 C4 and 16 C4, and differs from the amino acid sites 283 and 293 of 16 C4. Compared with "vaccine strains," there was a significant difference between the 50-59 years old age group when the NtAb titer of 16 C4 strain was 1:512-1:1024. Our results suggest that changes in the functional epitopes of NtAb caused by amino acid 283 and 293 loci in EV-A71 strains may affect the production of neutralizing antibodies.

Retrospective immunogenicity analysis of seasonal flu H3N2 vaccines recommended in the past ten years using immunized animal sera

BACKGROUND

Influenza A (H3N2) virus (A/H3N2) has complex antigenic evolution, resulting in frequent vaccine strain updates. We aimed to evaluate the protective effect of the vaccine strains on the circulating strains from past ten years and provide a basis for finding a broader and more efficient A/H3N2 vaccine strain.

METHODS

Eighty-four representative circulating A/H3N2 strains were selected from 65,791 deposited sequences in 2011-2020 and pseudotyped viruses were constructed with the VSV vector. We immunized guinea pigs with DNA vaccine containing the A/H3N2 components of the vaccine strains from 2011 to 2021 and tested neutralizing antibody against the pseudotyped viruses. We used a hierarchical clustering method to classify the eighty-four representative strains into different antigenic clusters. We also immunized animals with monovalent vaccine stock of the vaccine strains for the 2020-2021 and 2021-2022 seasons and tested neutralizing antibody against the pseudotyped viruses.

FINDINGS

The vaccine strains PE/09, VI/11 and TE/12 induced higher levels of neutralizing antibody against representative strains circulating in recommended year and the year immediately prior whereas vaccine strains HK/14, HK/19 and CA/20 induced poor neutralization against all representative strains. The representative strains were divided into five antigenic clusters (AgV), which were not identical to gene clades. The AgV5 strains were most difficult to be protected among the five clusters. Compared with single-dose immunization, three doses of monovalent vaccine stock (HK/19 or CA/20) could induce stronger and broader neutralizing antibodies against strains in each of the antigenic clusters.

INTERPRETATION

The protective effect of vaccine strains indicated that the accurate selection of A/H3N2 vaccine strains must remain a top priority. By increasing the frequency of immunization, stronger and broader neutralizing antibodies against strains in all antigenic clusters were induced, which provides direction for a new immunization strategy.

FUNDING

This work was supported by a grant from National Key R&D Program of China (No. 2021YFC2301700).

The influences of the East Asian Monsoon on the spatio-temporal pattern of seasonal influenza activity in China

Previous research has extensively studied the seasonalities of human influenza infections and the effect of specific climatic factors in different regions. However, there is limited understanding of the influences of monsoons. This study applied generalized additive model with monthly surveillance data from mainland China to explore the influences of the East Asian Monsoon on the spatio-temporal pattern of seasonal influenza in China. The results suggested two influenza active periods in northern China and three active periods in southern China. The study found that the northerly advancement of East Asian Summer Monsoon (EASM) influences the summer influenza spatio-temporal patterns in both southern and northern China. At the interannual scale, the north-south converse effect of EASM on influenza activity is mainly due to the converse effect of EASM on humidity and precipitation. Within the annual scale, influenza activity in southern China gradually reaches its maximum during the summer exacerbated by the northerly advancement of EASM. Furthermore, the winter epidemic in China is related to the low temperature and humidity influenced by the East Asian Winter Monsoon (EAWM). Moreover, the active period in transition season is related partially to the large rapid temperature change influenced by the transition of EAWM and EASM. Despite the delayed onset and instability, the climatic condition influenced by the East Asian Monsoon is one of the potential key drivers of influenza activity.

Transmission Patterns of Seasonal Influenza in China between 2010 and 2018

Background Understanding the transmission source, pattern, and mechanism of infectious diseases is essential for targeted prevention and control. Though it has been studied for many years, the detailed transmission patterns and drivers for the seasonal influenza epidemics in China remain elusive. Methods In this study, utilizing a suite of epidemiological and genetic approaches, we analyzed the updated province-level weekly influenza surveillance, sequence, climate, and demographic data between 1 April 2010 and 31 March 2018 from continental China, to characterize detailed transmission patterns and explore the potential initiating region and drivers of the seasonal influenza epidemics in China. Results An annual cycle for influenza A(H1N1)pdm09 and B and a semi-annual cycle for influenza A(H3N2) were confirmed. Overall, the seasonal influenza A(H3N2) virus caused more infection in China and dominated the summer season in the south. The summer season epidemics in southern China were likely initiated in the “Lingnan” region, which includes the three most southern provinces of Hainan, Guangxi, and Guangdong. Additionally, the regions in the south play more important seeding roles in maintaining the circulation of seasonal influenza in China. Though intense human mobility plays a role in the province-level transmission of influenza epidemics on a temporal scale, climate factors drive the spread of influenza epidemics on both the spatial and temporal scales. Conclusion The surveillance of seasonal influenza in the south, especially the “Lingnan” region in the summer, should be strengthened. More broadly, both the socioeconomic and climate factors contribute to the transmission of seasonal influenza in China. The patterns and mechanisms revealed in this study shed light on the precise forecasting, prevention, and control of seasonal influenza in China and worldwide.

Status and Challenges for Vaccination against Avian H9N2 Influenza Virus in China

In China, H9N2 avian influenza virus (AIV) has become widely prevalent in poultry, causing huge economic losses after secondary infection with other pathogens. Importantly, H9N2 AIV continuously infects humans, and its six internal genes frequently reassort with other influenza viruses to generate novel influenza viruses that infect humans, threatening public health. Inactivated whole-virus vaccines have been used to control H9N2 AIV in China for more than 20 years, and they can alleviate clinical symptoms after immunization, greatly reducing economic losses. However, H9N2 AIVs can still be isolated from immunized chickens and have recently become the main epidemic subtype. A more effective vaccine prevention strategy might be able to address the current situation. Herein, we analyze the current status and vaccination strategy against H9N2 AIV and summarize the progress in vaccine development to provide insight for better H9N2 prevention and control.

Continued evolution of H6 avian influenza viruses isolated from farms in China between 2014 and 2018

H6 avian influenza virus (AIV) is one of the most prevalent AIV subtypes in the world. Our previous studies have demonstrated that H6 AIVs isolated from live poultry markets pose a potential threat to human health. In recent years, increasing number of H6 AIVs has been constantly isolated from poultry farms. In order to understand the biological characteristics of H6 AIVs in the context of farms, here, we analyzed the phylogenetic relationships, antigenicity, replication in mice and receptor binding properties of H6 AIVs isolated from farms in China between 2014 and 2018. Phylogenetic analysis showed that 19 different genotypes were formed among 20 representative H6 viruses. Notably, the internal genes of these H6 viruses exhibited complicated relationships with different subtypes of AIVs worldwide, indicating that these viruses are the products of complex and frequent reassortment events. Antigenic analysis revealed that 13 viruses tested were divided into three antigenic groups. 10 viruses examined could all replicate in the respiratory organs of infected mice without prior adaptation. Receptor binding analysis demonstrated that some of the H6 AIVs bound to both α-2, 3-linked glycans (avian-type receptor) and α-2, 6-linked glycans (human-type receptor), thereby posing a potential threat to human health. Together, these findings revealed the prevalence, complicated genetic evolution, diverse antigenicity, and dual receptor binding specificity of H6 AIVs in the settings of poultry farms, which emphasize the importance to continuously monitor the evolution and biological properties of H6 AIVs in nature.

Antigenic characterization of influenza and SARS-CoV-2 viruses

Antigenic characterization of emerging and re-emerging viruses is necessary for the prevention of and response to outbreaks, evaluation of infection mechanisms, understanding of virus evolution, and selection of strains for vaccine development. Primary analytic methods, including enzyme-linked immunosorbent/lectin assays, hemagglutination inhibition, neuraminidase inhibition, micro-neutralization assays, and antigenic cartography, have been widely used in the field of influenza research. These techniques have been improved upon over time for increased analytical capacity, and some have been mobilized for the rapid characterization of the SARS-CoV-2 virus as well as its variants, facilitating the development of highly effective vaccines within 1 year of the initially reported outbreak. While great strides have been made for evaluating the antigenic properties of these viruses, multiple challenges prevent efficient vaccine strain selection and accurate assessment. For influenza, these barriers include the requirement for a large virus quantity to perform the assays, more than what can typically be provided by the clinical samples alone, cell- or egg-adapted mutations that can cause antigenic mismatch between the vaccine strain and circulating viruses, and up to a 6-month duration of vaccine development after vaccine strain selection, which allows viruses to continue evolving with potential for antigenic drift and, thus, antigenic mismatch between the vaccine strain and the emerging epidemic strain. SARS-CoV-2 characterization has faced similar challenges with the additional barrier of the need for facilities with high biosafety levels due to its infectious nature. In this study, we review the primary analytic methods used for antigenic characterization of influenza and SARS-CoV-2 and discuss the barriers of these methods and current developments for addressing these challenges.

Comparison of observer based methods for source localisation in complex networks

In recent years, research on methods for locating a source of spreading phenomena in complex networks has seen numerous advances. Such methods can be applied not only to searching for the "patient zero" in epidemics, but also finding the true sources of false or malicious messages circulating in the online social networks. Many methods for solving this problem have been established and tested in various circumstances. Yet, we still lack reviews that would include a direct comparison of efficiency of these methods. In this paper, we provide a thorough comparison of several observer-based methods for source localisation on complex networks. All methods use information about the exact time of spread arrival at a pre-selected group of vertices called observers. We investigate how the precision of the studied methods depends on the network topology, density of observers, infection rate, and observers' placement strategy. The direct comparison between methods allows for an informed choice of the methods for applications or further research. We find that the Pearson correlation based method and the method based on the analysis of multiple paths are the most effective in networks with synthetic or real topologies. The former method dominates when the infection rate is low; otherwise, the latter method takes over.

Birth cohort relative to an influenza A virus's antigenic cluster introduction drives patterns of children's antibody titers

An individual's antibody titers to influenza A strains are a result of the complicated interplay between infection history, cross-reactivity, immune waning, and other factors. It has been challenging to disentangle how population-level patterns of humoral immunity change as a function of age, calendar year, and birth cohort from cross-sectional data alone. We analyzed 1,589 longitudinal sera samples from 260 children across three studies in Nicaragua, 2006-16. Hemagglutination inhibition (HAI) titers were determined against four H3N2 strains, one H1N1 strain, and two H1N1pdm strains. We assessed temporal patterns of HAI titers using an age-period-cohort modeling framework. We found that titers against a given virus depended on calendar year of serum collection and birth cohort but not on age. Titer cohort patterns were better described by participants' ages relative to year of likely introduction of the virus's antigenic cluster than by age relative to year of strain introduction or by year of birth. These cohort effects may be driven by a decreasing likelihood of early-life infection after cluster introduction and by more broadly reactive antibodies at a young age. H3N2 and H1N1 viruses had qualitatively distinct cohort patterns, with cohort patterns of titers to specific H3N2 strains reaching their peak in children born 3 years prior to that virus's antigenic cluster introduction and with titers to H1N1 and H1N1pdm strains peaking for children born 1-2 years prior to cluster introduction but not being dramatically lower for older children. Ultimately, specific patterns of strain circulation and antigenic cluster introduction may drive population-level antibody titer patterns in children.

Univ-flu: A structure-based model of influenza A virus hemagglutinin for universal antigenic prediction

The rapid mutations on hemagglutinin (HA) of influenza A virus (IAV) can lead to significant antigenic variance and consequent immune mismatch of vaccine strains. Thus, rapid antigenicity evaluation is highly desired. The subtype-specific antigenicity models have been widely used for common subtypes such as H1 and H3. However, the continuous emerging of new IAV subtypes requires the construction of universal antigenic prediction model which could be applied on multiple IAV subtypes, including the emerging or re-emerging ones. In this study, we presented Univ-Flu, series structure-based universal models for HA antigenicity prediction. Initially, the universal antigenic regions were derived on multiple subtypes. Then, a radial shell structure combined with amino acid indexes were introduced to generate the new three-dimensional structure based descriptors, which could characterize the comprehensive physical–chemical property changes between two HA variants within or across different subtypes. Further, by combining with Random Forest classifier and different training datasets, Univ-Flu could achieve high prediction performances on intra-subtype (average AUC of 0.939), inter-subtype (average AUC of 0.771), and universal-subtype (AUC of 0.978) prediction, through independent test. Results illustrated that the designed descriptor could provide accurate universal antigenic description. Finally, the application on high-throughput antigenic coverage prediction for circulating strains showed that the Univ-Flu could screen out virus strains with high cross-protective spectrum, which could provide in-silico reference for vaccine recommendation.

Influenza A and B outbreaks differed in their associations with climate conditions in Shenzhen, China

Under the variant climate conditions in the transitional regions between tropics and subtropics, the impacts of climate factors on influenza subtypes have rarely been evaluated. With the available influenza A (Flu-A) and influenza B (Flu-B) outbreak data in Shenzhen, China, which is an excellent example of a transitional marine climate, the associations of multiple climate variables with these outbreaks were explored in this study. Daily laboratory-confirmed influenza virus and climate data were collected from 2009 to 2015. Potential impacts of daily mean/maximum/minimum temperatures (T/T_max/T_min), relative humidity (RH), wind velocity (V), and diurnal temperature range (DTR) were analyzed using the distributed lag nonlinear model (DLNM) and generalized additive model (GAM). Under its local climate partitions, Flu-A mainly prevailed in summer months (May to June), and a second peak appeared in early winter (December to January). Flu-B outbreaks usually occurred in transitional seasons, especially in autumn. Although low temperature caused an instant increase in both Flu-A and Flu-B risks, its effect could persist for up to 10 days for Flu-B and peak at 17 C (relative risk (RR) = 14.16, 95% CI: 7.46-26.88). For both subtypes, moderate-high temperature (28 C) had a significant but delayed effect on influenza, especially for Flu-A (RR = 26.20, 95% CI: 13.22-51.20). The Flu-A virus was sensitive to RH higher than 76%, while higher Flu-B risks were observed at both low (< 65%) and high (> 83%) humidity. Flu-A was active for a short term after exposure to large DTR (e.g., DTR = 10 C, RR = 12.45, 95% CI: 6.50-23.87), whereas Flu-B mainly circulated under stable temperatures. Although the overall wind speed in Shenzhen was low, moderate wind (2-3 m/s) was found to favor the outbreaks of both subtypes. This study revealed the thresholds of various climatic variables promoting influenza outbreaks, as well as the distinctions between the flu subtypes. These data can be helpful in predicting seasonal influenza outbreaks and minimizing the impacts, based on integrated forecast systems coupled with short-term climate models.

A statistical analysis of antigenic similarity among influenza A (H3N2) viruses

An accurate assessment of antigenic similarity between influenza viruses is important for vaccine strain recommendations and influenza surveillance. Due to the mechanisms that result in frequent changes in the antigenicities of strains, it is desirable to obtain an antigenic similarity measure that accounts for specific changes in strains that are of epidemiological importance in influenza. Empirically grounded statistical models best achieve this. In this study, an interpretable machine-learning model was developed using distinguishing features of antigenic variants to analyze antigenic similarity. The features comprised of cluster information, amino acid sequences located in known antigenic and receptor-binding sites of influenza A (H3N2). In order to assess validity of parameters, accuracy and relevance of model to vaccine effectiveness, the model was applied to influenza A (H3N2) viruses due to their abundant genetic data and epidemiological relevance to influenza surveillance. An application of the model revealed that all model parameters were statistically significant to determining antigenic similarity between strains. Furthermore, upon evaluating the model for predicting antigenic similarity between strains, it achieved 95% area under Receiver Operating Characteristic curve (AUC), 94% accuracy, 76% precision, 97% specificity, 68% sensitivity and a diagnostic odds ratio (DOR) of 83.19. Above all, the model was found to be strongly related to influenza vaccine effectiveness to indicate the correlation between vaccine effectiveness and antigenic similarity between vaccine and circulating strains in an epidemic. The study predicts probabilities of antigenic similarity and estimates changes in strains that lead to antigenic variants. A successful application of the methods presented in this study would complement the global efforts in influenza surveillance.

A Deep Learning Approach for Predicting Antigenic Variation of Influenza A H3N2

Modeling antigenic variation in influenza (flu) virus A H3N2 using amino acid sequences is a promising approach for improving the prediction accuracy of immune efficacy of vaccines and increasing the efficiency of vaccine screening. Antigenic drift and antigenic jump/shift, which arise from the accumulation of mutations with small or moderate effects and from a major, abrupt change with large effects on the surface antigen hemagglutinin (HA), respectively, are two types of antigenic variation that facilitate immune evasion of flu virus A and make it challenging to predict the antigenic properties of new viral strains. Despite considerable progress in modeling antigenic variation based on the amino acid sequences, few studies focus on the deep learning framework which could be most suitable to be applied to this task. Here, we propose a novel deep learning approach that incorporates a convolutional neural network (CNN) and bidirectional long-short-term memory (BLSTM) neural network to predict antigenic variation. In this approach, CNN extracts the complex local contexts of amino acids while the BLSTM neural network captures the long-distance sequence information. When compared to the existing methods, our deep learning approach achieves the overall highest prediction performance on the validation dataset, and more encouragingly, it achieves prediction agreements of 99.20% and 96.46% for the strains in the forthcoming year and in the next two years included in an existing set of chronological amino acid sequences, respectively. These results indicate that our deep learning approach is promising to be applied to antigenic variation prediction of flu virus A H3N2.

Antigenic cartography reveals complexities of genetic determinants that lead to antigenic differences among pandemic GII.4 noroviruses

Noroviruses are the predominant cause of acute gastroenteritis, with a single genotype (GII.4) responsible for the majority of infections. This prevalence is characterized by the periodic emergence of new variants that present substitutions at antigenic sites of the major structural protein (VP1), facilitating escape from herd immunity. Notably, the contribution of intravariant mutations to changes in antigenic properties is unknown. We performed a comprehensive antigenic analysis on a virus-like particle panel representing major chronological GII.4 variants to investigate diversification at the inter- and intravariant level. Immunoassays, neutralization data, and cartography analyses showed antigenic similarities between phylogenetically related variants, with major switches to antigenic properties observed over the evolution of GII.4 variants. Genetic analysis indicated that multiple coevolving amino acid changes-primarily at antigenic sites-are associated with the antigenic diversification of GII.4 variants. These data highlight complexities of the genetic determinants and provide a framework for the antigenic characterization of emerging GII.4 noroviruses.

SAS: A Platform of Spike Antigenicity for SARS-CoV-2

Since the outbreak of SARS-CoV-2, antigenicity concerns continue to linger with emerging mutants. As recent variants have shown decreased reactivity to previously determined monoclonal antibodies (mAbs) or sera, monitoring the antigenicity change of circulating mutants is urgently needed for vaccine effectiveness. Currently, antigenic comparison is mainly carried out by immuno-binding assays. Yet, an online predicting system is highly desirable to complement the targeted experimental tests from the perspective of time and cost. Here, we provided a platform of SAS (Spike protein Antigenicity for SARS-CoV-2), enabling predicting the resistant effect of emerging variants and the dynamic coverage of SARS-CoV-2 antibodies among circulating strains. When being compared to experimental results, SAS prediction obtained the consistency of 100% on 8 mAb-binding tests with detailed epitope covering mutational sites, and 80.3% on 223 anti-serum tests. Moreover, on the latest South Africa escaping strain (B.1.351), SAS predicted a significant resistance to reference strain at multiple mutated epitopes, agreeing well with the vaccine evaluation results. SAS enables auto-updating from GISAID, and the current version collects 867K GISAID strains, 15.4K unique spike (S) variants, and 28 validated and predicted epitope regions that include 339 antigenic sites. Together with the targeted immune-binding experiments, SAS may be helpful to reduce the experimental searching space, indicate the emergence and expansion of antigenic variants, and suggest the dynamic coverage of representative mAbs/vaccines among the latest circulating strains. SAS can be accessed at https://www.biosino.org/sas.

New framework for recombination and adaptive evolution analysis with application to the novel coronavirus SARS-CoV-2

The 2019 novel coronavirus (SARS-CoV-2) has spread rapidly worldwide and was declared a pandemic by the WHO in March 2020. The evolution of SARS-CoV-2, either in its natural reservoir or in the human population, is still unclear, but this knowledge is essential for effective prevention and control. We propose a new framework to systematically identify recombination events, excluding those due to noise and convergent evolution. We found that several recombination events occurred for SARS-CoV-2 before its transfer to humans, including a more recent recombination event in the receptor-binding domain. We also constructed a probabilistic mutation network to explore the diversity and evolution of SARS-CoV-2 after human infection. Clustering results show that the novel coronavirus has diverged into several clusters that cocirculate over time in various regions and that several mutations across the genome are fixed during transmission throughout the human population, including D614G in the S gene and two accompanied mutations in ORF1ab. Together, these findings suggest that SARS-CoV-2 experienced a complicated evolution process in the natural environment and point to its continuous adaptation to humans. The new framework proposed in this study can help our understanding of and response to other emerging pathogens.

The role of conformational epitopes in the evolutionary divergence of enterovirus D68 clades: A bioinformatics-based study

Enterovirus D68 (EV-D68), as one of the major pathogens of paediatric respiratory disease, has been widely spread in the population in recent years. As the basis of virus antigenicity, antigenic epitopes are essential to monitoring the transformation of virus antigenicity. However, there is a lack of systematic studies on the antigenic epitopes of EV-D68. In this study, a bioinformatics-based prediction algorithm for human enteroviruses was used to predict the conformational epitopes of EV-D68. The prediction results showed that the conformational epitopes of EV-D68 were clustered into three sites: site 1, site 2, and site 3. Site 1 was located in the "north rim" region of the canyon near the fivefold axis; site 2 was located in the "puff" region near the twofold axis; and site 3 consisted of two parts, one in the "knob" region on the south rim of the canyon and the other in the threefold axis region. The predicted epitopes overlapped highly with the binding regions of four reported monoclonal antibodies (mAbs), indicating that the predictions were highly reliable. Phylogenetic analysis showed that amino acid mutations in the epitopes of the VP1 BC loop, DE loop, C-terminus, and VP2 EF loop played a crucial role in the evolutionary divergence of EV-D68 clades/subclades and epidemics. This finding indicated that the VP1 BC loop, DE loop, C-terminus, and VP2 EF loop were the most important epitopes of EV-D68. Research on the epitopes of EV-D68 will contribute to outbreak surveillance and to the development of diagnostic reagents and recombinant vaccines.

Modeling the selective advantage of new amino acids on the hemagglutinin of H1N1 influenza viruses using their patient age distributions

In 2009, a new strain of H1N1 influenza A virus caused a pandemic, and its descendant strains are causing seasonal epidemics worldwide. Given the high mutation rate of influenza viruses, variant strains having different amino acids on hemagglutinin (HA) continuously emerge. To prepare vaccine strains for the next influenza seasons, it is an urgent task to predict which variants will be selected in the viral population. An analysis of 24,681 pairs of an amino acid sequence of HA of H1N1pdm2009 viruses and its patient age showed that the empirical fixation probability of new amino acids on HA significantly differed depending on their frequencies in the population, patient age distributions, and epitope flags. The selective advantage of a variant strain having a new amino acid was modeled by linear combinations of patients age distributions and epitope flags, and then the fixation probability of the new amino acid was modeled using Kimura’s formula for advantageous selection. The parameters of models were estimated from the sequence data and models were tested with four-fold cross validations. The frequency of new amino acids alone can achieve high sensitivity, specificity, and precision in predicting the fixation of a new amino acid of which frequency is more than 0.11. The estimated parameter suggested that viruses with a new amino acid having a frequency in the population higher than 0.11 have a significantly higher selective advantage compared to viruses with the old amino acid at the same position. The model considering the Z-value of patient age rank-sums of new amino acids predicted amino acid substitutions on HA with a sensitivity of 0.78, specificity of 0.86, and precision of 0.83, showing significant improvement compared to the constant selective advantage model, which used only the frequency of the amino acid. These results suggested that H1N1 viruses tend to be selected in the adult population, and frequency of viruses having new amino acids and their patient ages are useful to predict amino acid substitutions on HA.

Time series computational prediction of vaccines for influenza A H3N2 with recurrent neural networks

Influenza viruses are persistently threatening public health, causing annual epidemics and sporadic pandemics due to rapid viral evolution. Vaccines are used to prevent influenza infections but the composition of the influenza vaccines have to be updated regularly to ensure its efficacy. Computational tools and analyses have become increasingly important in guiding the process of vaccine selection. By constructing time-series training samples with splittings and embeddings, we develop a computational method for predicting suitable strains as the recommendation of the influenza vaccines using recurrent neural networks (RNNs). The Encoder-decoder architecture of RNN model enables us to perform sequence-to-sequence prediction. We employ this model to predict the prevalent sequence of the H3N2 viruses sampled from 2006 to 2017. The identity between our predicted sequence and recommended vaccines is greater than 98% and the Pepitope<0.2 indicates their antigenic similarity. The multi-step vaccine prediction further demonstrates the robustness of our method which achieves comparable results in contrast to single step prediction. The results show significant matches of the recommended vaccine strains to the circulating strains. We believe it would facilitate the process of vaccine selection and surveillance of seasonal influenza epidemics.

Type-IInterferon-Inducible SERTAD3 Inhibits Influenza A Virus Replication by Blocking the Assembly of Viral RNA Polymerase Complex

Influenza A virus (IAV) infection stimulates a type I interferon (IFN-I) response in host cells that exerts antiviral effects by inducing the expression of hundreds of IFN-stimulated genes (ISGs). However, most ISGs are poorly studied for their roles in the infection of IAV. Herein, we demonstrate that SERTA domain containing 3 (SERTAD3) has a significant inhibitory effect on IAV replication in vitro. More importantly, Sertad3^-/- mice develop more severe symptoms upon IAV infection. Mechanistically, we find SERTAD3 reduces IAV replication through interacting with viral polymerase basic protein 2 (PB2), polymerase basic protein 1 (PB1), and polymerase acidic protein (PA) to disrupt the formation of the RNA-dependent RNA polymerase (RdRp) complex. We further identify an 8-amino-acid peptide of SERTAD3 as a minimum interacting motif that can disrupt RdRp complex formation and inhibit IAV replication. Thus, our studies not only identify SERTAD3 as an antiviral ISG, but also provide the mechanism of potential application of SERTAD3-derived peptide in suppressing influenza replication.

Bioinformatics-based prediction of conformational epitopes for human parechovirus

Human parechoviruses (HPeVs) are human pathogens that usually cause diseases ranging from rash to neonatal sepsis in young children. HPeV1 and HPeV3 are the most frequently reported genotypes and their three-dimensional structures have been determined. However, there is a lack of systematic research on the antigenic epitopes of HPeVs, which are useful for understanding virus-receptor interactions, developing antiviral agents or molecular diagnostic tools, and monitoring antigenic evolution. Thus, we systematically predicted and compared the conformational epitopes of HPeV1 and HPeV3 using bioinformatics methods in the study. The results showed that both epitopes clustered into three sites (sites 1, 2 and 3). Site 1 was located on the "northern rim" near the fivefold vertex; site 2 was on the "puff"; and site 3 was divided into two parts, of which one was located on the "knob" and the other was close to the threefold vertex. The predicted epitopes highly overlapped with the reported antigenic epitopes, which indicated that the prediction results were accurate. Although the distribution positions of the epitopes of HPeV1 and HPeV3 were highly consistent, the residues varied largely and determined the genotypes. Three amino acid residues, VP3-91N, -92H and VP0-257S, were the key residues for monoclonal antibody (mAb) AM28 binding to HPeV1 and were also of great significance in distinguishing HPeV1 and HPeV3. We also found that two residues, VP1-85N and -87D, might affect the capability of mAb AT12-015 to bind to HPeV3.

Bioinformatics-based prediction of conformational epitopes for Enterovirus A71 and Coxsackievirus A16

Enterovirus A71 (EV-A71), Coxsackievirus A16 (CV-A16) and CV-A10 are the major causative agents of hand, foot and mouth disease (HFMD). The conformational epitopes play a vital role in monitoring the antigenic evolution, predicting dominant strains and preparing vaccines. In this study, we employed a Bioinformatics-based algorithm to predict the conformational epitopes of EV-A71 and CV-A16 and compared with that of CV-A10. Prediction results revealed that the distribution patterns of conformational epitopes of EV-A71 and CV-A16 were similar to that of CV-A10 and their epitopes likewise consisted of three sites: site 1 (on the "north rim" of the canyon around the fivefold vertex), site 2 (on the "puff") and site 3 (one part was in the "knob" and the other was near the threefold vertex). The reported epitopes highly overlapped with our predicted epitopes indicating the predicted results were reliable. These data suggested that three-site distribution pattern may be the basic distribution role of epitopes on the enteroviruses capsids. Our prediction results of EV-A71 and CV-A16 can provide essential information for monitoring the antigenic evolution of enterovirus.

FluPhenotype-a one-stop platform for early warnings of the influenza A virus

MOTIVATION

Newly emerging influenza viruses keep challenging global public health. To evaluate the potential risk of the viruses, it is critical to rapidly determine the phenotypes of the viruses, including the antigenicity, host, virulence and drug resistance.

RESULTS

Here, we built FluPhenotype, a one-stop platform to rapidly determinate the phenotypes of the influenza A viruses. The input of FluPhenotype is the complete or partial genomic/protein sequences of the influenza A viruses. The output presents five types of information about the viruses: (i) sequence annotation including the gene and protein names as well as the open reading frames, (ii) potential hosts and human-adaptation-associated amino acid markers, (iii) antigenic and genetic relationships with the vaccine strains of different HA subtypes, (iv) mammalian virulence-related amino acid markers and (v) drug resistance-related amino acid markers. FluPhenotype will be a useful bioinformatic tool for surveillance and early warnings of the newly emerging influenza A viruses.

AVAILABILITY AND IMPLEMENTATION

It is publicly available from: http://www.computationalbiology.cn : 18888/IVEW.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

A Mutation Network Method for Transmission Analysis of Human Influenza H3N2

Characterizing the spatial transmission pattern is critical for better surveillance and control of human influenza. Here, we propose a mutation network framework that utilizes network theory to study the transmission of human influenza H3N2. On the basis of the mutation network, the transmission analysis captured the circulation pattern from a global simulation of human influenza H3N2. Furthermore, this method was applied to explore, in detail, the transmission patterns within Europe, the United States, and China, revealing the regional spread of human influenza H3N2. The mutation network framework proposed here could facilitate the understanding, surveillance, and control of other infectious diseases.

Convolutional Neural Network Based Approach to in Silico Non-Anticipating Prediction of Antigenic Distance for Influenza Virus

Evaluation of the antigenic similarity degree between the strains of the influenza virus is highly important for vaccine production. The conventional method used to measure such a degree is related to performing the immunological assays of hemagglutinin inhibition. Namely, the antigenic distance between two strains is calculated on the basis of HI assays. Usually, such distances are visualized by using some kind of antigenic cartography method. The known drawback of the HI assay is that it is rather time-consuming and expensive. In this paper, we propose a novel approach for antigenic distance approximation based on deep learning in the feature spaces induced by hemagglutinin protein sequences and Convolutional Neural Networks (CNNs). To apply a CNN to compare the protein sequences, we utilize the encoding based on the physical and chemical characteristics of amino acids. By varying (hyper)parameters of the CNN architecture design, we find the most robust network. Further, we provide insight into the relationship between approximated antigenic distance and antigenicity by evaluating the network on the HI assay database for the H1N1 subtype. The results indicate that the best-trained network gives a high-precision approximation for the ground-truth antigenic distances, and can be used as a good exploratory tool in practical tasks.

A benchmark dataset of protein antigens for antigenicity measurement

Antigenicity measurement plays a fundamental role in vaccine design, which requires antigen selection from a large number of mutants. To augment traditional cross-reactivity experiments, computational approaches for predicting the antigenic distance between multiple protein antigens are highly valuable. The performance of in silico models relies heavily on large-scale benchmark datasets, which are scattered among public databases and published articles or reports. Here, we present the first benchmark dataset of protein antigens with experimental evidence to guide in silico antigenicity calculations. This dataset includes (1) standard haemagglutination-inhibition (HI) tests for 3,867 influenza A/H3N2 strain pairs, (2) standard HI tests for 559 influenza virus B strain pairs, and (3) neutralization titres derived from 1,073 Dengue virus strain pairs. All of these datasets were collated and annotated with experimentally validated antigenicity relationships as well as sequence information for the corresponding protein antigens. We anticipate that this work will provide a benchmark dataset for in silico antigenicity prediction that could be further used to assist in epidemic surveillance and therapeutic vaccine design for viruses with variable antigenicity.

Implementing sequence-based antigenic distance calculation into immunological shape space model

BACKGROUND

In 2009, a novel influenza vaccine was distributed worldwide to combat the H1N1 influenza "swine flu" pandemic. However, antibodies induced by the vaccine display differences in their specificity and cross-reactivity dependent on pre-existing immunity. Here, we present a computational model that can capture the effect of pre-existing immunity on influenza vaccine responses. The model predicts the region of the virus hemagglutinin (HA) protein targeted by antibodies after vaccination as well as the level of cross-reactivity induced by the vaccine. We tested our model by simulating a scenario similar to the 2009 pandemic vaccine and compared the results to antibody binding data obtained from human subjects vaccinated with the monovalent 2009 H1N1 influenza vaccine.

RESULTS

We found that both specificity and cross-reactivity of the antibodies induced by the 2009 H1N1 influenza HA protein were affected by the viral strain the individual was originally exposed. Specifically, the level of antigenic relatedness between the original exposure HA antigen and the 2009 HA protein affected antigenic-site immunodominance. Moreover, antibody cross-reactivity was increased when the individual's pre-existing immunity was specific to an HA protein antigenically distinct from the 2009 pandemic strain. Comparison of simulation data with antibody binding data from human serum samples demonstrated qualitative and quantitative similarities between the model and real-life immune responses to the 2009 vaccine.

CONCLUSION

We provide a novel method to evaluate expected outcomes in antibody specificity and cross-reactivity after influenza vaccination in individuals with different influenza HA antigen exposure histories. The model produced similar outcomes as what has been previously reported in humans after receiving the 2009 influenza pandemic vaccine. Our results suggest that differences in cross-reactivity after influenza vaccination should be expected in individuals with different exposure histories.

The complex associations of climate variability with seasonal influenza A and B virus transmission in subtropical Shanghai, China

Most previous studies focused on the association between climate variables and seasonal influenza activity in tropical or temperate zones, little is known about the associations in different influenza types in subtropical China. The study aimed to explore the associations of multiple climate variables with influenza A (Flu-A) and B virus (Flu-B) transmissions in Shanghai, China. Weekly influenza virus and climate data (mean temperature (MeanT), diurnal temperature range (DTR), relative humidity (RH) and wind velocity (Wv)) were collected between June 2012 and December 2018. Generalized linear models (GLMs), distributed lag non-linear models (DLNMs) and regression tree models were developed to assess such associations. MeanT exerted the peaking risk of Flu-A at 1.4 °C (2-weeks' cumulative relative risk (RR): 14.88, 95% confidence interval (CI): 8.67-23.31) and 25.8 °C (RR: 12.21, 95%CI: 6.64-19.83), Flu-B had the peak at 1.4 °C (RR: 26.44, 95%CI: 11.52-51.86). The highest RR of Flu-A was 23.05 (95%CI: 5.12-88.45) at DTR of 15.8 °C, that of Flu-B was 38.25 (95%CI: 15.82-87.61) at 3.2 °C. RH of 51.5% had the highest RR of Flu-A (9.98, 95%CI: 4.03-26.28) and Flu-B (4.63, 95%CI: 1.95-11.27). Wv of 3.5 m/s exerted the peaking RR of Flu-A (7.48, 95%CI: 2.73-30.04) and Flu-B (7.87, 95%CI: 5.53-11.91). DTR ≥ 12 °C and MeanT <22 °C were the key drivers for Flu-A and Flu-B, separately. The study found complex non-linear relationships between climate variability and different influenza types in Shanghai. We suggest the careful use of meteorological variables in influenza prediction in subtropical regions, considering such complex associations, which may facilitate government and health authorities to better minimize the impacts of seasonal influenza.

Co-circulation and persistence of multiple A/H3N2 influenza variants in China

The spread of influenza A/H3N2 variants possessing the hemagglutinin 121 K mutation and the unexpectedly high incidence of influenza in the 2017–2018 northern hemisphere influenza season have raised serious concerns about the next pandemic. We summarized the national surveillance data of seasonal influenza in China and identified marked differences in influenza epidemics between northern and southern China, particularly the predominating subtype and the presence of an additional summer peak in southern China. Notably, a minor spring peak of influenza caused by a different virus subtype was also observed. We also revealed that the 3C.2a lineage was dominant from the summer of 2015 to the end of the 2015–2016 peak season in China, after which the 3C.2a2 lineage predominated despite the importation and co-circulation of the 121 K variants of 3C.2a1 and 3C.2a3 lineages at the global level. Finally, an analysis based on genetic distances revealed a delay in A/H3N2 vaccine strain update. Overall, our results highlight the complicated circulation pattern of seasonal influenza in China and the necessity for a timely vaccine strain update worldwide.

Exploring the limitations of biophysical propensity scales coupled with machine learning for protein sequence analysis

Machine learning (ML) is ubiquitous in bioinformatics, due to its versatility. One of the most crucial aspects to consider while training a ML model is to carefully select the optimal feature encoding for the problem at hand. Biophysical propensity scales are widely adopted in structural bioinformatics because they describe amino acids properties that are intuitively relevant for many structural and functional aspects of proteins, and are thus commonly used as input features for ML methods. In this paper we reproduce three classical structural bioinformatics prediction tasks to investigate the main assumptions about the use of propensity scales as input features for ML methods. We investigate their usefulness with different randomization experiments and we show that their effectiveness varies among the ML methods used and the tasks. We show that while linear methods are more dependent on the feature encoding, the specific biophysical meaning of the features is less relevant for non-linear methods. Moreover, we show that even among linear ML methods, the simpler one-hot encoding can surprisingly outperform the “biologically meaningful” scales. We also show that feature selection performed with non-linear ML methods may not be able to distinguish between randomized and “real” propensity scales by properly prioritizing to the latter. Finally, we show that learning problem-specific embeddings could be a simple, assumptions-free and optimal way to perform feature learning/engineering for structural bioinformatics tasks.

A context-free encoding scheme of protein sequences for predicting antigenicity of diverse influenza A viruses

BACKGROUND

The evolution of influenza A viruses leads to the antigenic changes. Serological diagnosis of the antigenicity is usually labor-intensive, time-consuming and not suitable for early-stage detection. Computational prediction of the antigenic relationship between emerging and old strains of influenza viruses using viral sequences can facilitate large-scale antigenic characterization, especially for those viruses requiring high biosafety facilities, such as H5 and H7 influenza A viruses. However, most computational models require carefully designed subtype-specific features, thereby being restricted to only one subtype.

METHODS

In this paper, we propose a Context-FreeEncoding Scheme (CFreeEnS) for pairs of protein sequences, which encodes a protein sequence dataset into a numeric matrix and then feeds the matrix into a downstream machine learning model. CFreeEnS is not only free from subtype-specific selected features but also able to improve the accuracy of predicting the antigenicity of influenza. Since CFreeEnS is subtype-free, it is applicable to predicting the antigenicity of diverse influenza subtypes, hopefully saving the biologists from conducting serological assays for highly pathogenic strains.

RESULTS

The accuracy of prediction on each subtype tested (A/H1N1, A/H3N2, A/H5N1, A/H9N2) is over 85%, and can be as high as 91.5%. This outperforms existing methods that use carefully designed subtype-specific features. Furthermore, we tested the CFreeEnS on the combined dataset of the four subtypes. The accuracy reaches 84.6%, much higher than the best performance 75.1% reported by other subtype-free models, i.e. regional band-based model and residue-based model, for predicting the antigenicity of influenza. Also, we investigate the performance of CFreeEnS when the model is trained and tested on different subtypes (i.e. transfer learning). The prediction accuracy using CFreeEnS is 84.3% when the model is trained on the A/H1N1 dataset and tested on the A/H5N1, better than the 75.2% using a regional band-based model.

CONCLUSIONS

The CFreeEnS not only improves the prediction of antigenicity on datasets with only one subtype but also outperforms existing methods when tested on a combined dataset with four subtypes of influenza viruses.

Predicting antigenic variants of H1N1 influenza virus based on epidemics and pandemics using a stacking model

H1N1 is the earliest emerging subtype of influenza A viruses with available genomic sequences, has caused several pandemics and seasonal epidemics, resulting in millions of deaths and enormous economic losses. Timely determination of new antigenic variants is crucial for the vaccine selection and flu prevention. In this study, we chronologically divided the H1N1 strains into several periods in terms of the epidemics and pandemics. Computational models have been constructed to predict antigenic variants based on epidemic and pandemic periods. By sequence analysis, we demonstrated the diverse mutation patterns of HA1 protein on different periods and that an individual model built upon each period can not represent the variations of H1N1 virus. A stacking model was established for the prediction of antigenic variants, combining all the variation patterns across periods, which would help assess a new influenza strain’s antigenicity. Three different feature extraction methods, i.e. residue-based, regional band-based and epitope region-based, were applied on the stacking model to verify its feasibility and robustness. The results showed the capability of determining antigenic variants prediction with accuracy as high as 0.908 which performed better than any of the single models. The prediction performance using the stacking model indicates clear distinctions of mutation patterns and antigenicity between epidemic and pandemic strains. It would also facilitate rapid determination of antigenic variants and influenza surveillance.

Using an innovative method to develop the threshold of seasonal influenza epidemic in China

Background

Proper early warning thresholds for defining seasonal influenza epidemics are crucial for timely engagement of intervention strategies, but are currently not well established in China. We propose a novel moving logistic regression method (MLRM) to determine epidemic thresholds and validate them with the Chinese influenza surveillance data.

Methods

For each province, historical epidemic waves are formed as weekly percentages of laboratory-confirmed patients among all clinically diagnosed influenza cases. For each epidemic curve that is approximately symmetric, a series of logistic curves are fitted to increasing temporal range of the epidemic, and the threshold is determined based on the best-fitting logistic curve.

Results

Using surveillance data of seasonal influenza collected during 2010–2014 in 30 provinces of China, we screened 153 epidemic waves and identified 100 as approximately symmetric; and 85 of the 100 waves were satisfactorily fitted. Compared to two published approaches, the MLRM identified lower thresholds of seasonal influenza epidemics, leading to about three weeks earlier detection of onset and about four weeks later detection of closure of the epidemics. The potential misclassification proportion of influenza epidemic waves was 6% for the MLRM, comparable to that for the two published approaches.

Conclusions

The MLRM offers an alternative to existing methods for defining early warning thresholds for the surveillance of seasonal influenza, and can be readily generalized to other countries and other infectious agents. The thresholds we identified can be used for early detection of future influenza epidemics in China.

Evolution-informed forecasting of seasonal influenza A (H3N2)

Interpandemic or seasonal influenza A, currently subtypes H3N2 and H1N1, exacts an enormous annual burden both in terms of human health and economic impact. Incidence prediction ahead of season remains a challenge largely because of the virus' antigenic evolution. We propose a forecasting approach that incorporates evolutionary change into a mechanistic epidemiological model. The proposed models are simple enough that their parameters can be estimated from retrospective surveillance data. These models link amino acid sequences of hemagglutinin epitopes with a transmission model for seasonal H3N2 influenza, also informed by H1N1 levels. With a monthly time series of H3N2 incidence in the United States for more than 10 years, we demonstrate the feasibility of skillful prediction for total cases ahead of season, with a tendency to underpredict monthly peak epidemic size, and an accurate real-time forecast for the 2016/2017 influenza season.

CE-BLAST makes it possible to compute antigenic similarity for newly emerging pathogens

Major challenges in vaccine development include rapidly selecting or designing immunogens for raising cross-protective immunity against different intra- or inter-subtypic pathogens, especially for the newly emerging varieties. Here we propose a computational method, Conformational Epitope (CE)-BLAST, for calculating the antigenic similarity among different pathogens with stable and high performance, which is independent of the prior binding-assay information, unlike the currently available models that heavily rely on the historical experimental data. Tool validation incorporates influenza-related experimental data sufficient for stability and reliability determination. Application to dengue-related data demonstrates high harmonization between the computed clusters and the experimental serological data, undetectable by classical grouping. CE-BLAST identifies the potential cross-reactive epitope between the recent zika pathogen and the dengue virus, precisely corroborated by experimental data. The high performance of the pathogens without the experimental binding data suggests the potential utility of CE-BLAST to rapidly design cross-protective vaccines or promptly determine the efficacy of the currently marketed vaccine against emerging pathogens, which are the critical factors for containing emerging disease outbreaks.

Deploying digital health data to optimize influenza surveillance at national and local scales

The surveillance of influenza activity is critical to early detection of epidemics and pandemics and the design of disease control strategies. Case reporting through a voluntary network of sentinel physicians is a commonly used method of passive surveillance for monitoring rates of influenza-like illness (ILI) worldwide. Despite its ubiquity, little attention has been given to the processes underlying the observation, collection, and spatial aggregation of sentinel surveillance data, and its subsequent effects on epidemiological understanding. We harnessed the high specificity of diagnosis codes in medical claims from a database that represented 2.5 billion visits from upwards of 120,000 United States healthcare providers each year. Among influenza seasons from 2002-2009 and the 2009 pandemic, we simulated limitations of sentinel surveillance systems such as low coverage and coarse spatial resolution, and performed Bayesian inference to probe the robustness of ecological inference and spatial prediction of disease burden. Our models suggest that a number of socio-environmental factors, in addition to local population interactions, state-specific health policies, as well as sampling effort may be responsible for the spatial patterns in U.S. sentinel ILI surveillance. In addition, we find that biases related to spatial aggregation were accentuated among areas with more heterogeneous disease risk, and sentinel systems designed with fixed reporting locations across seasons provided robust inference and prediction. With the growing availability of health-associated big data worldwide, our results suggest mechanisms for optimizing digital data streams to complement traditional surveillance in developed settings and enhance surveillance opportunities in developing countries.

Influenza contributes substantially to global morbidity and mortality each year, and epidemiological surveillance for influenza is typically conducted by sentinel physicians and health care providers recruited to report cases of influenza-like illness. While population coverage and representativeness, and geographic distribution are considered during sentinel provider recruitment, systems cannot always achieve these standards due to the administrative burdens of data collection. We present spatial estimates of influenza disease burden across United States counties by leveraging the volume and fine spatial resolution of medical claims data, and existing socio-environmental hypotheses about the determinants of influenza disease disease burden. Using medical claims as a testbed, this study adds to literature on the optimization of surveillance system design by considering conditions of limited reporting and spatial aggregation. We highlight the importance of considering sampling biases and reporting locations when interpreting surveillance data, and suggest that local mobility and regional policies may be critical to understanding the spatial distribution of reported influenza-like illness.

Predictive Modeling of Influenza Shows the Promise of Applied Evolutionary Biology

Seasonal influenza is controlled through vaccination campaigns. Evolution of influenza virus antigens means that vaccines must be updated to match novel strains, and vaccine effectiveness depends on the ability of scientists to predict nearly a year in advance which influenza variants will dominate in upcoming seasons. In this review, we highlight a promising new surveillance tool: predictive models. Based on data-sharing and close collaboration between the World Health Organization and academic scientists, these models use surveillance data to make quantitative predictions regarding influenza evolution. Predictive models demonstrate the potential of applied evolutionary biology to improve public health and disease control. We review the state of influenza predictive modeling and discuss next steps and recommendations to ensure that these models deliver upon their considerable biomedical promise.

Incidence Prediction for the 2017-2018 Influenza Season in the United States with an Evolution-informed Model

Introduction:

Seasonal influenza is responsible for a high disease burden in the United States and worldwide. Predicting outbreak size in advance can contribute to the timely control of seasonal influenza by informing health care and vaccination planning.

Methods:

Recently, a process-based model was developed for forecasting incidence dynamics ahead of the season, with the approach validated by several statistical criteria, including an accurate real-time prediction for the past 2016-2017 influenza season before it started.

Results:

Based on this model and data up to June 2017, a forecast for the upcoming 2017-2018 influenza season is presented here, indicating an above-average, moderately severe, outbreak dominated by the H3N2 subtype.

Discussion:

The prediction is consistent with surveillance data so far, which already indicate the predominance of H3N2. The forecast for the upcoming 2017-2018 influenza season reinforces the importance of the on-going vaccination campaign.

Computationally Optimized Broadly Reactive Hemagglutinin Elicits Hemagglutination Inhibition Antibodies against a Panel of H3N2 Influenza Virus Cocirculating Variants

Each influenza season, a set of wild-type viruses, representing one H1N1, one H3N2, and one to two influenza B isolates, are selected for inclusion in the annual seasonal influenza vaccine. In order to develop broadly reactive subtype-specific influenza vaccines, a methodology called computationally optimized broadly reactive antigens (COBRA) was used to design novel hemagglutinin (HA) vaccine immunogens. COBRA technology was effectively used to design HA immunogens that elicited antibodies that neutralized H5N1 and H1N1 isolates. In this report, the development and characterization of 17 prototype H3N2 COBRA HA proteins were screened in mice and ferrets for the elicitation of antibodies with HA inhibition (HAI) activity against human seasonal H3N2 viruses that were isolated over the last 48 years. The most effective COBRA HA vaccine regimens elicited antibodies with broader HAI activity against a panel of H3N2 viruses than wild-type H3 HA vaccines. The top leading COBRA HA candidates were tested against cocirculating variants. These variants were not efficiently detected by antibodies elicited by the wild-type HA from viruses selected as the vaccine candidates. The T-11 COBRA HA vaccine elicited antibodies with HAI and neutralization activity against all cocirculating variants from 2004 to 2007. This is the first report demonstrating broader breadth of vaccine-induced antibodies against cocirculating H3N2 strains compared to the wild-type HA antigens that were represented in commercial influenza vaccines.IMPORTANCE There is a need for an improved influenza vaccine that elicits immune responses that recognize a broader number of influenza virus strains to prevent infection and transmission. Using the COBRA approach, a set of vaccines against influenza viruses in the H3N2 subtype was tested for the ability to elicit antibodies that neutralize virus infection against not only historical vaccine strains of H3N2 but also a set of cocirculating variants that circulated between 2004 and 2007. Three of the H3N2 COBRA vaccines recognized all of the cocirculating strains during this era, but the chosen wild-type vaccine strains were not able to elicit antibodies with HAI activity against these cocirculating strains. Therefore, the COBRA vaccines have the ability to elicit protective antibodies against not only the dominant vaccine strains but also minor circulating strains that can evolve into the dominant vaccine strains in the future.