Limits...
Predicting Avian Influenza Co-Infection with H5N1 and H9N2 in Northern Egypt

View Article: PubMed Central - PubMed

ABSTRACT

Human outbreaks with avian influenza have been, so far, constrained by poor viral adaptation to non-avian hosts. This could be overcome via co-infection, whereby two strains share genetic material, allowing new hybrid strains to emerge. Identifying areas where co-infection is most likely can help target spaces for increased surveillance. Ecological niche modeling using remotely-sensed data can be used for this purpose. H5N1 and H9N2 influenza subtypes are endemic in Egyptian poultry. From 2006 to 2015, over 20,000 poultry and wild birds were tested at farms and live bird markets. Using ecological niche modeling we identified environmental, behavioral, and population characteristics of H5N1 and H9N2 niches within Egypt. Niches differed markedly by subtype. The subtype niches were combined to model co-infection potential with known occurrences used for validation. The distance to live bird markets was a strong predictor of co-infection. Using only single-subtype influenza outbreaks and publicly available ecological data, we identified areas of co-infection potential with high accuracy (area under the receiver operating characteristic (ROC) curve (AUC) 0.991).

No MeSH data available.


Conceptual diagram showing the creation of five niche models and two derived models, with rectangles indicating data used for model testing or validation.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC5036719&req=5

ijerph-13-00886-f002: Conceptual diagram showing the creation of five niche models and two derived models, with rectangles indicating data used for model testing or validation.

Mentions: We employed ENM to identify a set of environmental conditions associated with virus persistence and spread [48]. ENM examines the environmental data at locations of outbreaks and creates a model of the virus’ niche, which is evaluated and improved using randomly selected background points within the study area [49]. A large number of modeling algorithms are available with varying strengths and weaknesses, including regression-based models and machine learning methods [50,51]. We chose the maximum entropy modeling algorithm, which is robust to multicollinearity among input variables, and has been found to provide more conservative estimates of species niches than other common algorithms, including the Genetic Algorithm for Rule-set Production (GARP) and bioclimate analysis and prediction system (BIOCLIM), even with small sample sizes [52,53,54,55]. Presence-only models were created for outbreaks between 2007 and 2015 in Northern Egypt according to virus subtype, and co-infection potential maps were derived from them (see Figure 2).


Predicting Avian Influenza Co-Infection with H5N1 and H9N2 in Northern Egypt
Conceptual diagram showing the creation of five niche models and two derived models, with rectangles indicating data used for model testing or validation.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC5036719&req=5

ijerph-13-00886-f002: Conceptual diagram showing the creation of five niche models and two derived models, with rectangles indicating data used for model testing or validation.
Mentions: We employed ENM to identify a set of environmental conditions associated with virus persistence and spread [48]. ENM examines the environmental data at locations of outbreaks and creates a model of the virus’ niche, which is evaluated and improved using randomly selected background points within the study area [49]. A large number of modeling algorithms are available with varying strengths and weaknesses, including regression-based models and machine learning methods [50,51]. We chose the maximum entropy modeling algorithm, which is robust to multicollinearity among input variables, and has been found to provide more conservative estimates of species niches than other common algorithms, including the Genetic Algorithm for Rule-set Production (GARP) and bioclimate analysis and prediction system (BIOCLIM), even with small sample sizes [52,53,54,55]. Presence-only models were created for outbreaks between 2007 and 2015 in Northern Egypt according to virus subtype, and co-infection potential maps were derived from them (see Figure 2).

View Article: PubMed Central - PubMed

ABSTRACT

Human outbreaks with avian influenza have been, so far, constrained by poor viral adaptation to non-avian hosts. This could be overcome via co-infection, whereby two strains share genetic material, allowing new hybrid strains to emerge. Identifying areas where co-infection is most likely can help target spaces for increased surveillance. Ecological niche modeling using remotely-sensed data can be used for this purpose. H5N1 and H9N2 influenza subtypes are endemic in Egyptian poultry. From 2006 to 2015, over 20,000 poultry and wild birds were tested at farms and live bird markets. Using ecological niche modeling we identified environmental, behavioral, and population characteristics of H5N1 and H9N2 niches within Egypt. Niches differed markedly by subtype. The subtype niches were combined to model co-infection potential with known occurrences used for validation. The distance to live bird markets was a strong predictor of co-infection. Using only single-subtype influenza outbreaks and publicly available ecological data, we identified areas of co-infection potential with high accuracy (area under the receiver operating characteristic (ROC) curve (AUC) 0.991).

No MeSH data available.