Mining the NCBI Influenza Sequence Database: adaptive grouping of BLAST results using precalculated neighbor indexing

The Influenza Virus Resource and other Virus Variation Resources at NCBI provide enhanced visualization web tools for exploratory analysis for influenza sequence data. Despite the improvements in data analysis, the initial data retrieval remains unsophisticated, frequently producing huge and imbalanced datasets due to the large number of identical and nearly-identical sequences in the database. We propose a data mining algorithm to organize reported sequences into groups based on their relatedness to the query sequence and to each other. The algorithm uses BLAST to find database sequences related to the query. Neighbor lists precalculated from pairwise BLAST alignments between database sequences are used to organize results in groups of nearly-identical and strongly related sequences. We propose to use a non-symmetric dissimilarity measure well crafted for dealing with sequences of different length (fragments). A balanced and representative data set produced by this tool can be used for further analysis, i.e. multiple sequence alignment and phylogenetic trees. The algorithm is implemented for protein coding sequences and is being integrated with the NCBI Influenza Virus Resource.