Local protein structures to bridge sequence-structure knowledge

Abstract

Protein sequences can be classified based on their structure similarity and/or common evolutionary origin called structural class. Information on structural class is readily available, easing the protein structure and protein function probing. SCOP and CATH are two prominent classification schemes used to assign the structural class of proteins. Both schemes determine the structural class manually base on known protein tertiary structures. However, the quantity of known protein sequences is growing exponentially with respect to the quantity of known tertiary proteins structures. Although SCOP and CATH are examples of well-established databases that contain more reliable information of structural class, yet the lack of known structural class of protein due to the laborious wet-lab experimental routine limits the high-throughput structural class assignment. The fact that this is a tedious and time-consuming manually-determined method has further limited the structural class assignment. As a consequence, the assignment of structural class by computational method suffers from the arbitrated statistical infer-ence. Thus, this study aims to provide a structural class prediction method that can acquire the knowledge of local protein structures, derived from known excessive primary sequences, in order to produce high-throughput sequence-structure class assignment instead of the laborious experimental based method. This structural class prediction method is termed as SVM-LpsSCPred.