Predicting the potential for zoonotic transmission and host associations for novel viruses

Abstract

Host-virus associations have co-evolved under ecological and evolutionary selection pressures that shape cross-species transmission and spillover to humans. Observed virus-host associations provide relevant context for newly discovered wildlife viruses to assess knowledge gaps in host-range and estimate pathways for potential human infection. Using models to predict virus-host networks, we predicted the likelihood of humans as hosts for 513 newly discovered viruses detected by large-scale wildlife surveillance at high-risk animal-human interfaces in Africa, Asia, and Latin America. Predictions indicated that novel coronaviruses are likely to infect a greater number of host species than viruses from other families. Our models further characterize novel viruses through prioritization scores and directly inform surveillance targets to identify host ranges for newly discovered viruses.

Fig. 1. Modeling workflow.

The figure shows the modeling procedure and methods implemented in the study. Orange dots represent a known virus in the observed ($G_c$) and predicted networks ($G_{predicted}$), blue dots represent novel viruses in the predicted network ($G_{predicted}$). Virus-host networks: $G_c$, represents a unipartite observed network of known zoonotic and non-zoonotic viruses with nodes representing viruses and edges representing shared hosts. $G_{predicted}$represents the predicted unipartite network generated after predicting possible linkages between 531 novel viruses (blue) and known viruses. The node size is proportional to the betweenness centrality.

Fig. 2. Predicting missing links between virus-host communities.

Distribution shapes of degree (a) and betweenness centrality (b) for the observed and predicted network. Degree distributions for virus families in observed and predicted networks are shown in e and f. Similarly, shapes of betweenness centrality for virus families in observed and predicted networks are shown in i and j. Right panels show boxplots for novel virus families describing degree (c), betweenness centrality (d), eigenvector centrality (g), and clustering based on the predicted network formed by the binary prediction model (h).

Fig. 3. Prioritization metrics for novel viruses to understand zoonotic risk.

Top ten and bottom five newly discovered viruses from six virus families (a–d) with the virus prioritization scores based on multiclass model predictions. Annotations show the score and support represented by the number of human links predicted.

Fig. 4. Surveillance targets for novel coronaviruses based on predicted sharing of hosts with known viruses.

The red color represents the evidence of species in the taxonomic family (cumulative probability) with darker red color indicating a higher number of species occurrences from taxonomical families adjusted by model predicted probability. a shows clustering of novel coronaviruses by the host, and b focuses on novel coronaviruses found in bats. Clustering is based on the Bray-Curtis dissimilarity index.