Aim: Both bacterial infection and viral infection involve a large number of protein-protein
interactions (PPIs) between a pathogen and its target host.
Background: So far, many computational methods have focused on predicting PPIs within the same
species rather than PPIs across different species.
Methods: From the extensive analysis of PPIs between Yersinia pestis bacteria and humans, we recently
discovered an interesting relation; a linear relation between amino acid composition and sequence
length was observed in many proteins involved in PPIs. We have built a support vector machine (SVM)
model, which predicts PPIs between human and bacteria using two feature types derived from the relation.
The two feature types used in the SVM are the amino acid composition group (AACG) and the
difference in amino acid composition between host and pathogen proteins.
Results: The SVM model achieved high performance in predicting bacteria-human PPIs. The model
showed an accuracy of 96%, sensitivity of 94%, and specificity of 98% in predicting PPIs between humans
and Yersinia pestis, in which there is a strong relation between amino acid composition and sequence
length. The SVM model was also tested in predicting PPIs between human and viruses, which
include Ebola, HCV, and SARS-CoV-2, and showed a good performance.
Conclusion: The feature types identified in our study are simple yet powerful in predicting pathogenhuman
PPIs. Although preliminary, our method will be useful for finding unknown target host proteins
or pathogen proteins and designing in vitro or in vivo experiments.