2.2 Identification of domain-domain interactions
Identification was done only for those protein structures whose domains
interact with each other. To define interacting domains, 5-5 rule was
used, which states that interacting domains have at least five residue
contacts within 5Å. The distance criterion was adopted using our
in-house PIC software. Further, at least five interactions arising from
residues of domains were considered that are at least six residues apart
to consider a short linker connecting two domains, which would include
linkers of varying lengths. The classification of DDIs in monomeric
multi-domain proteins to obligate and non-obligate (here permanent and
transient) ones were done using NOXclass. This tool is a SVM classifier
which is based on the physicochemical properties of the interface. For
this study, we used the parameters which showed highest accuracy using
multi-stage SVM. Necessary PDB manipulations were done using pdb-tools.
To get a cutoff to define the interaction as obligate or non-obligate,
this was tested on Block et al. dataset and tried to match the
accuracy of prediction of NOXclass with different cutoffs, resulting in
a cutoff of 70% to distinguish the interaction as obligate and
non-obligate. We also used multiple structures of proteins to check
large structural deviations (>2Å) using MUSTANG, and
reclassified the interaction, wherever needed, based on literature.