Protein Sequence-Structure Alignment Using 3D-HMM

Tertiary structure of proteins provides valuable information on their biochemical functions. But experimental structure determination is still labor-intensive, in spite of technological advances. Therefore, rapid and accurate protein structure prediction method is not only interesting from a theoretical point of view but also practically useful. Currently, homology modeling, or comparative modeling, is the most reliable structure prediction method and widely used. It is applicable when structure of at least one homolog of the target protein is solved, and predicts 3D coordinates by aligning the target sequence and the template structure. It is noted that the accuracy of homology model depends largely on the accuracy of target-template alignment. But the alignment accuracy of conventional alignment algorithms declines sharply when the sequence identity between the target and the template proteins is lower than 45% [4]. Here we introduce a novel sequence-structure alignment method to improve the alignment accuracy between distantly related proteins. It is based on hidden Markov model (HMM), which have been successfully applied to many bioinformatics problems. The distinguishing feature of our method is the explicit consideration of 3D coordinates of each amino acid residue. Each state of the HMM has a coordinate in 3D space. Alexandrov and Gerstein first introduced such idea in the field of protein structure classification, and abbreviated it as 3D-HMM [1]. In this study, we adapted 3D-HMM for the protein sequence-structure alignment problem and tested its ability in alignment accuracy. These spatial coordinates allow us to introduce structure-based transition probabilities such as restriction on the distance between two consecutive Cα atoms. Moreover, an alignment path generated by 3D-HMM has explicit 3D coordinates. Thus, energy of the model can be directly calculated without additional model building procedure. This can greatly enhance the speed of model quality evaluation when we have a large number of alternative models.