The short silence

The paper: A. Birmingham et al., "3' UTR seed matches, but not overall identity, are associated with RNAi off-targets," Nat Methods, 3:199—204, 2006. (Cited in 94 papers) The finding: In 2006, Devin Leake and colleagues at Dharmacon Research wanted to determine the accuracy of current algorithms for predicting siRNA mismatch events used in silencing. To do t

Edyta Zielinska
Apr 1, 2008

The paper:

A. Birmingham et al., "3' UTR seed matches, but not overall identity, are associated with RNAi off-targets," Nat Methods, 3:199—204, 2006. (Cited in 94 papers)

The finding:

In 2006, Devin Leake and colleagues at Dharmacon Research wanted to determine the accuracy of current algorithms for predicting siRNA mismatch events used in silencing. To do this they generated a database of mismatched siRNA events by experimental gene array for three genes and compared them to those predicted by the Smith-Waterman algorithm. They found that only a small region of the siRNA was important for silencing, and that the algorithm could not predict hundreds of mismatch events.

The surprise:

Only a small nucleotide "seed" portion of the siRNA (5—6 nucleotides) was important for binding to the gene targeted for silencing, which increased the potential for nonspecific binding. In addition, only the seed region "defines the specificity" of silencing events,...

The significance:

The article demonstrated that off-target matches could occur more frequently and wouldn't be predicted by widely used (but now irrelevant) algorithms such as the Basic Local Alignment Search Tool, which used sequences similar to the full-length siRNA, says Patzel.

The follow-up:

Leake and his group developed a chemical modification of the siRNA that didn't alter the binding region but "decreased the off-target events up to 90%," says Leake. The new product, ON-TARGET Plus, from Thermo Scientific Dharmacon is more expensive than an unmodified siRNA, says Patzel, but much more specific.

Algorithm vs. experimental results
10,752 — Off-target events predicted by the Smith-Waterman algorithm
347 — Off-target events experimentally validated by researchers
23 — Events common to both data sets