Algorithm


It is well known that final outcome of siRNA efficacy is the contribution of efficacy gain at each step of RNAi pathway from loading of guide strand into RISC target accessibility, and cleavage efficiency. The principle of our method is to design siRNAs, which gain efficacy at various steps of RNAi pathways and at last incorporate the mismatch effect with target.

Dataset information:-
Dataset comprising 2431 siRNAs (training, and test1) was taken from (Huesken, et al., 2005). For further assessment of our model we include another test2 having 419 siRNA taken from other literatures (Harborth, et al., 2003; Khvorova, et al., 2003; Reynolds, et al., 2004; Ui-Tei, et al., 2004; Vickers, et al., 2003). We have implemented SVMlight (Thorsten, 1999) for regression models development. Experimental data having silencing effect of single and double-nucleotide mismatch between siRNAs and targets were taken from and Huand, et al., 2009 and Dahlgren, et al., 2008 respectively. All models were trained and tested by five-fold cross validation on training datasets.

Development of model for efficacy prediction:-
This tool having SVM model developed using feature of mino-,di-, tri-nucleotide frequency, binary pattern and target site accessibility inplemented on RBF kernel. The performance of our model is comparable with other well-known methods when tested on independent data set.

Efficacy of siRNA due to base mismatch:-
This study involves generation of mutation in antisense siRNA of 19 nt on every position with specific nucleotide. Efficacies of these mutated siRNAs were predicted using our SVM model. However the mutation generated in siRNA when bound with target sequence cause mismatch and hence reduced the silencing efficiency. To find out overall possible efficacy of mutated siRNA we incorporate the effect of mismatch taken from experimental data.

Mismatch efficacy incorporating both position and identity of nucleotide: Initially we generated the single mutation in siRNA. The repression changes effected by position of mismatch and identity of mismatch between siRNA:target is taken from experimental data (Huang et al., 2005). Figure 6 shows change in efficacy due to mismatch at each position. Therefore a mismatch efficacy is calculated by deduced in efficacy due to mismatch from predicted efficacy.
Mismatch efficacy incorporating only position effect: When effects of mismatch with position as well as identity were not found from these experimental data, we only considered the position specific mismatch effect. For position specific mismatch average effect of that position was considered. Mismatch Efficacy=predicted efficacy-reduced efficacy (due to mismatch).

  
b