Poster #RP102
A new method for predicting and maximizing the protein sequence alignment accuracy
Minho Lee*, Chan-seok Jeong*, Dongsup Kim*
*Korea Advanced Institute of Science and Technology, Daejeon, Korea
Recently, comparative protein structure modeling has gained a great performance. To ensure the accurate comparative modeling, finding the accurate sequence alignment is critical. However, in most fold recognition-methods, the main research focus has been on how to recognize the most similar template to a target protein. In this work, we developed a new method to predict the protein sequence alignment accuracy. For alignment accuracy measure, we used both shift score and maxsub score. The profile-profile alignment between a query protein and a template protein of length n in our fold library was transformed into an input vector of length n +2. The SVRs were trained to predict the alignment accuracy scores between structural alignment and profile-profile alignment. The performance was assessed by Pearson correlation coefficient. The trained SVR predicted the maxsub scores with the correlation of 0.86. Our method can be used to select the alignment parameters and option that would give the best alignment for each different pair of proteins.