Optimal and adaptive pairwise DNA sequence correlation analysis in natural gradient

Besides scoring function methods, correlation analysis is another index to indicate homology of the sequences. In this paper, the statistical properties of correlation between two comparing sequences were analyzed. Natural gradient method was proposed to extract the most uncorrelated components of sequences in an adaptive way and the optimal Performance Index (PI) was to measure the correlation between two sequences. This can be accomplished in a computational effective way by a fastest natural gradient falling algorithm. Result of the method is compared to conventional scoring methods and it shows that PI values are consistent with their BLAST identity rate. Natural gradient method can bypass the selection of scoring function and gap penalty parameters, which is a key and difficult problem in traditional sequence alignment algorithms, such as BLAST and FASTA etc.