Analysis of Antisense Oligodeoxynucleotides Based on Sequence Motif Content to Predict Effectiveness

As interest in antisense oligodeoxynucleotide technology continues to grow, the search for a computational means of predicting the efficacy particular oligos has increased as well. Antisense technology allows the targeted down-regulation of an mRNA by application of a complimentary (antisense) phosphorothioate DNA oligo, typically about 20 nt in length. Much remains unknown about which particular target on the mRNA will be most effective, and it is costly and time consuming to perform in vivo screenings of multiple oligos to determine the site [1]. This has led to the interest in a computational solution to determining effective oligos/sites. At present, most studies have focused on two primary means of predicting RNA site (oligo) efficacy. One approach examines the predicted secondary structure of the target RNA for sites that might be most accessible to oligo binding [2]. A related but distinct approach calculates binding energies between oligo and target, with more recent studies taking disruption energies of secondary structures into account [3, 4]. These studies have demonstrated some successes, but have not as yet achieved the level of accuracy that would be expected if structural and energetic considerations were the only factors acting in vivo to determine antisense efficacy. The work of Tu and colleagues indicated that the short tetranucleotide TCCC, when present in the oligo, greatly improves its chances of having a significant effect on RNA expression [5]. We have further studied correlations between short (3-mer, 4-mer) motifs and oligo efficacy, and have determined there are several motifs which have a strong correlation with antisense action on an RNA target. We have utilized two distinct machine-learning methods to demonstrate that motif-based analysis can be used to predict antisense efficacy with similar or better accuracy as that of the other types of studies.