Profile-profile alignment: a powerful tool for protein structure prediction

The problem of computing the tertiary structure of a protein from a given amino acid sequence has been a major subject of bioinformatics research during the last decade. Many different approaches have been taken to tackle the problem, the most successful of which are based on searching databases to identify a similar amino acid sequence in the PDB and using the corresponding structure as a template for modeling the structure of the query sequence. An important advance for the evaluation of sequence similarity in this context has been the use of a frequency profile that represents a part of the protein sequence space close to the query sequence instead of the query sequence itself. In this paper, we present a further extension of this principle by using profiles instead of the template sequences, also. We show that, by using our newly developed scoring model, the profile-profile alignment approach is able to significantly outperform current state of the art methods like PSI-BLAST, HMMs, or threading methods in a fold recognition setup. This is especially interesting since we show that it holds for closely related sequences as well as for very distantly related ones.