Analysis of Clustering Fragmented Protein Bond Angles

The desire for accurate protein prediction algorithms has been a hallmark of computational biology achievements. Still, better algorithms and methodologies can achieve even greater success with implication across a diverse range of biological and medicinal fields such as protein function inference. Accurate prediction methods rely heavily on sequence similarity, however structure is more evolutionary conserved, i.e. structure is an alternate characteristic for ancestral relationships between proteins. The premise of this work is that similar structural features will be clustered together, which may show a unique amino acid and secondary structure (SS) distribution, which can be, incorporated into HMMs for SS prediction and protein function inference algorithms. With structural-evolutionary relationship in mind, I propose a methodology for ‘structure’ based SS prediction methods using HMM and k-mean and fuzzy k -means fragmented protein clusters. When fragment distributions were incorporated into HMMs, the average accuracy increased by 1 percent but showed an increase in accuracy of up to 13 percent for particular sequences. The HMM results were not so promising, however the clustering of protein structure fragments by C-alphas bond angles shows to be a useful length-independent metric for inferring functional relationships between proteins.