Integration of shared-pathogen networks and machine learning reveals the key aspects of zoonoses and predicts mammalian reservoirs

Abstract

Diseases that spread to humans from animals, zoonoses, pose major threats to human health. Identifying animal reservoirs of zoonoses and predicting future outbreaks are increasingly important to human health and well-being and economic stability, particularly where research and resources are limited. Here, we integrate complex networks and machine learning approaches to develop a new approach to identifying reservoirs. An exhaustive dataset of mammal-pathogen interactions was transformed into networks where hosts are linked via their shared pathogens. We present a methodology for identifying important and influential hosts in these networks. Ensemble models linking network characteristics with phylogeny and life-history traits are then employed to predict those key hosts and quantify the roles they undertake in pathogen transmission. Our models reveal drivers explaining host importance and demonstrate how these drivers vary by pathogen taxa. Host importance is further integrated into ensemble models to predict reservoirs of zoonoses of various pathogen taxa and quantify the extent of pathogen sharing between humans and mammals. We establish predictors of reservoirs of zoonoses, showcasing host influence to be a key factor in determining these reservoirs. Finally, we provide new insight into the determinants of zoonosis-sharing, and contrast these determinants across major pathogen taxa.

Predicting mammalian hosts in which novel coronaviruses can be generated

“…knowledge of all of these factors could greatly add to our ability to assess ‘likeliness’ of homologous recombination, however, the available data are too limited for a study with the breadth of interactions we characterise here, and hence were unable to be included. Research effort, centring mainly on coronaviruses found in humans and their domesticated animals, can lead to overestimation of the potential of coronaviruses to recombine in frequently studied mammals, such as lab rodents that were excluded from the results reported here (similar to previous work 17 ), and significantly, domesticated pigs and cats that we have found to be important recombination host species of coronaviruses. We believe that this limitation is partially mitigated; first, methodologically, the effect of research effort has been limited by capturing similarities from our three points of view (virus, host and network) and multiple characteristics therein.…”

“…Topological features of ecological networks have been successfully utilised to enhance our understanding of pathogen sharing 17 , 18 , disease emergence and spill-over events 19 , and as means to predict missing links in host–pathogen networks 20 – 22 . Here, we capture this topology, and relations between coronaviruses and hosts in our network, by means of node (coronaviruses and hosts) embeddings using DeepWalk 23 —a deep learning method that has been successfully used to predict drug-target 24 and IncRNA-disease associations 25 .…”

Research perspectives on animal health in the era of artificial intelligence

“…For example, phylogenetic models offer an interesting perspective for identifying environmental bacterial strains with high infectious potentiality [ 35 ], or for predicting the existence of putative host reservoirs or vectors [ 36 ]. The analysis of pathogen sharing among hosts has been used to classify the potential reservoirs of zoonotic diseases using machine learning [ 37 ]. The analysis of pathogen genomes can also be used to identify genotypes of animal pathogens that are more likely to infect humans [ 38 ].…”

Data proliferation, reconciliation, and synthesis in viral ecology

“…As an illustrative example, we describe the analytical hurdles of working with host-virus association data , a format that characterizes the global virome as a bipartite network of hosts and viruses, with pairs connected by observed potential for infection. Recent studies highlight the central role for these data in efforts to understand viral macroecology and evolution (Carlson et al 2019, Dallas et al 2019, Albery et al 2020), to predict zoonotic emergence risk (Han et al 2015, 2016, Olival et al 2017, Wardeh et al 2020), and to anticipate the impacts of global environmental change on infectious disease (Carlson et al 2020, Gibb et al 2020, Johnson et al 2020). Several bespoke datasets have been compiled to address these questions, and as interest in these topics has grown, so has the fragmentation of total knowledge across those datasets.…”

“…Interactions are stored as a geographic edgelist, where each carrier and cargo can also have locality information; additional metadata include the number of sequences in GenBank and related publications. EID2’s dynamic web interface (currently available through download on a limited query-by-query basis which researchers often manually bind or by personal correspondence with data curators) contains information encompassing 4,799 mammal “carrier” species and 70,614 microparasite or macroparasite “cargo” species, of which 9,605 are viruses (Wardeh et al 2020). However, many researchers continue to use the static, open release of EID2 from a 2015 data paper (Wardeh et al 2015), which we focus on here for comparative purposes as a stable version of the database available to the community of practice.…”