A simple clustering technique for the design of rotamer libraries based on pairs of consecutive Residues

The protein side chain packing is a well-known NPhard computational problem and of prime relevance to the study of protein structure-function relationship and protein design. Central to the solution of this problem is the design of rotamer libraries. These libraries have improved with respect to the coverage of torsion angles, at the cost of becoming larger and more complex. However, larger libraries imply a larger size of the algorithm’s search space. To address this issue, we propose a method for designing rotamer libraries which considers pairs of consecutive amino acids, the resulting library of pairs leads to an exponential reduction of the size of the search space the algorithm needs to analyze. The proposed approach is based on the k-modes clustering algorithm to generate the library. The proposed method starts with a set of target protein structures and a desired coverage level of dihedral angles in a target dataset, then for each pair of consecutive residues in the target set, generates a set of pairs of rotamers that attains the prespecified coverage level. We show that our proposed method generates good achievable accuracy when built on a set of 149 structures and tested over two sets of 65 and 373 protein structures.