Abstract

Principles that govern protein folding still remain elusive. Given the huge sequence space, it is reasonable to assume that sequences follow a particular pattern to attain one of the folds already defined in the relatively small structural space. In this study, we have used protein structure networks at different interaction strengths of non-covalent interactions (Imin) (Brinda & Vishveshwara, 2005; Kannan & Vishveshwara, 1999), to identify patterns that can distinguish a native protein from decoy/modelled structures. This is a rigorous extension of an earlier study performed at Imin ⩾ 0% (Chatterjee, Bhattacharyya et al., 2012). Network properties such as the size of the largest cluster (SLClu), largest k-2 communities (ComSk2) and clustering coefficients (CCoe) are analysed for 5422 native structures and 29543 decoy/modelled structures. Steeper transition profile of the native structures as a function of Imin is consistently observed (see Figure) . The network properties generated at different Imin and main-chain hydrogen bonds (MHB) are integrated into support vector machine to build a classifier, giving an accuracy of 94.11%. The uniqueness of the protein structures through side-chain interactions are captured by the network parameters, while MHB represents the backbone packing. Quality predictions for the recently concluded CASP 10 predicted models are also performed using the model with the selected ones showing RMSD values < 2.5 Å with respect to the native structures. Amongst the network properties, ComSk2 is maximally able to capture the transition properties of the structures. Importance of ComSk2 has earlier been implicated to capture the percolating behaviour of a protein structure (Deb & Vishveshwara, 2009). Overall, a robust classifier is obtained, and patterns specific to native structures have been analysed and discussed. The study highlights the importance of side-chain interactions at different Imins, along with backbone level interactions.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call