siRNA off-target search

Designing highly effective and gene-specific short interfering RNA (siRNA) sequences is crucial for any biological applications involving RNA interference (RNAi). A critical requirement for applying RNAi process in therapeutic applications is the ability to predict and to avoid side effect interactions with unintended transcripts (messager RNA, or mRNA). In this paper, we propose a flexible framework for detecting siRNA off-target effects. The framework can also be extended with minor changes to other applications such as selecting PCR primers or microarray nucleotide probes.Based on the framework, we have developed and implemented a new homology sequence search program -- siRNA Off-target Search (SOS). SOS uses a hybrid, q-gram based approach, combining two filtering techniques using overlapping and non-overlapping q-grams. This approach considers three types of imperfect matches based on biological experiments: G:U wobbles, mismatches, and bulges. The three main improvements over existing methods are: 1) introduce a more general cost model (an affine bulge cost model) for siRNA-mRNA off-target alignment; 2) use separate searches for alignments with and without bulges that enables efficient discovery of potential off-target candidates in the filtration phase; and 3) achieve better performance, in terms of speed and recall/precision, than BLAST in detecting potential siRNA off-targets.