HMI-PRED 2.0 searches for potential host-microbe interactions by screening known protein-protein interactions as templates. The chart above shows the template statistics. HMI-PRED 2.0 templates are pulled from RCSB PDB, where at least one chain is a human protein. Endogeneous interaction templates are those with two human proteins, where either one of chains can be mimicked by microbial proteins to interact with the other chain. Exogeneous interaction templates are those with one human and one non-human chain, and microbes are scanned for possible interactions with the human chain. In this case, we assume microbes can only mimic non-human chains, and can only interact with human chains.
To efficiently search for potential host-microbe interactions, templates are clustered by the identity of protein-protein interactions and the structural similarities. Representative templates of the clusters (one representative per cluster) are scanned for potential mimicry by the input microbe protein.

Microbial proteins are structurally aligned and matched evolutionarily conserved hot spots against all representative template chains (only non-human in exogeneous templates). The complementary chains of successfully mimicked (aligned) template chains are then computationally docked to predict host-microbe interactions that are not only structurally well-aligned, but also energetically favorable. Hot spot conservation level indicates the degree of interface hot spots matching between microbial and mimicked chains. The interface hot spots are computationally predicted for each mimic chains. If a hot spot residue from a mimic chain is structurally aligned to the microbe protein, The diagrams above show the numbers of microbial proteins and hot spot conservation levels of predicted HMIs.
HMI-PRED 2.0 has scanned 9.7 billion microbe-template pairs, where 10840752 pairs passed structural alignment. 1200125 HMIs have been predicted. The box plots below are to show distributions of structural alignment scores and docking scores for the predicted HMIs. (Last updated May 28, 2023, 5:59 a.m.)