Fast and Complete Search of siRNA Off-target Sequences

Smith-Waterman alignment algorithm is favored in search for siRNA off-target instead of the BLAST algorithm, because BLAST tends to overlook some significant homologous sequences, especially when they are short (21 nt~27 nt). Smith-Waterman algorithm, however, suffers from its own shortcomings, especially its inefficiency in searching through a large sequence database. This paper presents a two-phase homology search strategy that preserves the strength of SmithWaterman alignment algorithm while shortening its running time. In the first phase of this algorithm, selected siRNA sequence is divided into multiple mutually disjoint substrings, each of which is used to scan the sequence database for perfect matches against other genes. Only the sequences that have perfect match to substrings (of a given siRNA) are kept for the second phase. The second phase is the bona fide Smith-Waterman procedure. During this phase, the algorithm only checks the local vicinity sequences where a substring lands on a perfect match. This two-phased arrangement of the algorithm significantly improves the efficiency of the original Smith-Waterman algorithm by concentrating the search on localized regions instead of the whole genome sequence.