(A) In the set of predicted interactions (PRE) a majority of host proteins interacted with a low number of pathogen proteins and vice versa. Such a result was confirmed by the set of external interaction data (EXT) as well as the combined set of host-parasite interactions (PRE+EXT). (B) Utilizing a network of human protein-protein interactions the enrichment of targeted human host proteins as a function of their number of interaction partners was determined. Considering all three sets of host-pathogen interactions separately highly connected host proteins appeared to be prime targets of the parasite.

(A) In a bipartite matrix a parasite protein was linked to a pathway if the corresponding targeted proteins were enriched in the given pathway (FDR<0.05, Fisher's exact test). Utilizing the combined set of host-parasite interactions the matrix was composed of 83 parasite proteins that interacted with proteins in 117 pathways. In addition, a large fraction of pathways had >95% of their proteins being expressed in liver. Specifically, a cluster of prominent signaling pathways emerged that were largely under the control of parasite specific chaperone proteins. In (B) all interactions between parasite chaperones and host proteins that appeared in the cluster of pathways were mapped. Significantly, such a network revolves around parasite proteins PF08_0054 and PFI0875w that garnered most of the host-parasite interactions and largely interfered with proteins of the TNF pathway.

(A) In a network of human protein interactions the mean number of pathways a given human protein is involved in increased with the number of interaction partners (inset). The enrichment of such targeted pathways was determined as a function of their number of interaction partners. Utilizing predicted (PRE), external (EXT) and both sets (PRE+EXT) of host-parasite interactions, highly connected, targeted proteins appeared to be increasingly involved in such pathways. (B) A low value of the pathway participation coefficient indicated that the interaction partners of a protein in a human protein interaction network reached many different pathways and vice versa. Considering all human proteins that were involved in pathways, the majority of proteins have low pathway participation coefficients. Considering targeted proteins in all sets of host-parasite interactions the initially observed trend was significantly reinforced in all cases (Wilcoxon rank-sum test, P<10⁻¹⁰).

(A) As a positive training set I utilized 1,112 interactions between the human host and the parasite P. falciparum that have been previously inferred from protein structures. A non-interacting training set of equal size was constructed by randomly sampling pairs of human and parasite proteins that did not appear in the positive training set. Applying the random forest algorithm pairs of proteins that were incorrectly classified as interacting were discarded, and counts of correctly classified pairs were updated. If the number of pairs in the negative set that were sampled at least 3 times was roughly the size of the positive training set, the procedure terminated. Otherwise, the negative set of was filled with randomly sampled protein pairs until positive and negative training sets had the same size again. Previously described steps were repeated until the procedure finally terminated, providing a negative set of 1,136 non-interacting pairs. (B) Applying the random forest algorithm the training sets allowed for a true positive rate TPR = 78.9% and a false positive rate FPR = 4.7% (dashed lines) in a pronounced ROC curve. In a test-retest analysis, the classifier was trained on ⅔^rd of randomly picked training data. Its performance was tested on the remaining ⅓^rd, indicating that the proposed method was largely robust to noise.

A candidate interaction was identified in the web of human interactions if an interacting human protein had an ortholog in P. falciparum. Analogously, a potential interaction occurred if a protein in the parasite interaction network had a human ortholog. In a set of experimentally determined host-parasite interactions a host-parasite interaction was found, if the interacting proteins had homologs in the corresponding organism. The combination of all sources provided a total of 106,317 candidate interactions. The quality of the predicted interactions was assessed using the random forest algorithm, a machine learning method that classifies interactions as a function of the corresponding protein's sequences. Subsequently, 12,406 interactions thus obtained were filtered if they involved parasite proteins that were exported or carried molecular characteristics, enabling them to interact with the human host. While this step provided 6,229 interactions only host-parasite interactions were accounted for that occurred between proteins expressed in the parasitic merozoite stage and human red blood cell as well as in the sporozoite stage and liver cells. Accordingly, partially overlapping sets of 378 and 2,044 interactions were obtained, pooling a total of 2,244 predicted host-parasite protein interactions. In the last step predicted interactions were combined with external data such as structurally inferred and experimentally obtained interactions, providing a set of 3,322 combined host-parasite interactions.