ICON is an iterative, traveling salesman problem (TSP)-based clustering method for identifying near-native protein structures from an ensemble of conformers. Clustering of structures is carried out in the dihedral angle space and follows the TSP implementation of the rigorous global rearrangement approach OREO (DiMaggio et al., 2008). The method consists of an iterative procedure, which aims at eliminating clusters of structures at each iteration that are unlikely to be of similar fold to the native, based on a statistical analysis of cluster density and average spherical radius (see figure below). This procedure is carried out for 10 iterations, or until the number of conformers in the working set is less than 50% of the number of conformers in the original ensemble.

At the end of the iterative clustering procedure, the structures most likely to be close to the putative native structure are selected. To do this, the medoids of the densest clusters at the end of the final stage are collected. For each of these cluster mediods, novel Cα-Cα and centroid-centroid distance-dependent, high-resolution force fields (Rajgaria et al., 2006; Rajgaria et al., 2007) are implemented. These force fields aim to isolate native and near-native folds of a protein as lower energy structures, compared to structures further away from the native structure. The five cluster medoids with the lowest energies from each force field are selected as the top structures.

For more detailed information about how ICON works, please refer to Publications.

Click here to submit a job.

Flow sheet representing the ICON algorithm (reproduced from Figure 1 in Subramani et al., 2009)

Subramani, A.; DiMaggio, P. A.; Floudas, C. A. Selecting High Quality Protein Structures from Diverse Conformational Ensembles. Biophys. J. 2009, 97, 728-1736.
Find article on Google Scholar | Import citation into BibTeX, EndNote