Ecology of Middle East respiratory syndrome coronavirus, 2012-2020: A machine learning modelling analysis

Abstract

The ongoing enzootic circulation of the Middle East respiratory syndrome coronavirus (MERS-CoV) in the Middle East and North Africa is increasingly raising the concern about the possibility of its recombination with other human-adapted coronaviruses, particularly the pandemic SARS-CoV-2. We aim to provide an updated picture about ecological niches of MERS-CoV and associated socio-environmental drivers. Based on 356 confirmed MERS cases with animal contact reported to the WHO and 63 records of animal infections collected from the literature as of 30 May 2020, we assessed ecological niches of MERS-CoV using an ensemble model integrating three machine learning algorithms. With a high predictive accuracy (area under receiver operating characteristic curve = 91.66% in test data), the ensemble model estimated that ecologically suitable areas span over the Middle East, South Asia and the whole North Africa, much wider than the range of reported locally infected MERS cases and test-positive animal samples. Ecological suitability for MERS-CoV was significantly associated with high levels of bareland coverage (relative contribution = 30.06%), population density (7.28%), average temperature (6.48%) and camel density (6.20%). Future surveillance and intervention programs should target the high-risk populations and regions informed by updated quantitative analyses.

Keywords: MERS-CoV; Middle East respiratory syndrome; machine learning; predicted map; risk factors.

FIGURE 1

The flow chart for the selection of case districts that have evidence for presence of MERS‐CoV, that is human MERS cases with animal contact and PCR‐positive animal samples

FIGURE 2

Ecological assessment for MERS‐CoV using the ensemble model. The spatial distribution of case districts with either human MERS cases with animal contact or MERS‐CoV‐positive animal samples is shown in panel a. The spatial distribution of camel densities by averaging over 100 imputed data sets is shown in panel b. Ensemble‐model‐predicted risks of MERS‐CoV presence were mapped in panel c, and the uncertainty levels (standard deviation of predicted risk values over the 100 imputed data sets) were mapped in panel d

FIGURE 3

Contributions of key socioenvironmental factors to ecological suitability of MERS‐CoV identified by the BRT model. (a) The relative contributions of 20 variables included in the BRT model. Points represent the mean values, and bars represent the 95% confidence intervals. High (≥10%) and moderate (≥5%, <10%) relative contributions are coloured in red and blue, respectively. (b) Risk curves of the top nine influential factors. A higher risk value corresponds to a better ecological suitability for the virus. The distribution of each factor in the data is shown as the histograms in orange. (c) Box plots showing difference in the distribution of the top nine contributors between case districts (n = 117) and the pool of all control districts (n = 1820). Extreme values are removed. The centre line represents the median, and the box represent the inter‐quartile range. P‐values are based on Wilcoxon rank sum test except two‐sample t‐test is used for bareland coverage