One of the most powerful ways to predict a protein's unknown structure/function is to use of comparison with other proteins that already have structures/functions verified. This approach is so called "Comparative analysis". We aim to develop a computational approach to understand the protein structure and function relationship. Comparative analysis using evolutionarily related proteins would be a effective approach for our work. To achieve this goal, we develop and improve the bioinformatics algorithms such as sequence alignment and remote homology detection.


  In order to do the most effective and accurate prediction, we're doing :


1. Sequence alignment quality improvement
     The most important step for the prediction. If sequence alignments become more accurate, more precise putative template proteins can be found.

  r1.PNG

 

2. Search for the remote homolog
     A protein's remotely homologous proteins are defined as those that have low sequence identity with a qurey protein(a protein of which structure/function we want to identify), but have similar structures or functions with a query.
     we are aming at finding those remote homologs with better performance than commonly being used technique such as 'PSI-BLAST'.

 

r2.png

 

 

3. Algorithms and classifiers design
     we're taking use of various kinds of bioinformatics-related databases, and mathematical/statistical principles to do our research. Also for a better protein's structure/function prediction, we're using hyper-spatial machine learning techniques like SVM.