Accurate prediction of sRNA targets plays a key role in determining sRNA functions. Here we introduced two mathematical models, sRNATargetNB and sRNATargetSVM, for prediction of sRNA targets using Naïve Bayes method and support vector machines, respectively. The training dataset was composed of 46 positive samples (real sRNA-targets interaction) and 86 negative samples (no interaction between sRNA and targets). The leave-one-out cross-validation (LOOCV) classification accuracy was 91.67% for sRNATargetNB, and 100.00% for sRNATargetSVM. To evaluate the performance of the models, an independent test dataset was used, which contained 22 positive samples and 1700 randomly-generated negative samples. The results showed that the classification accuracy, sensitivity, and specificity were 93.03%, 40.90%, and 93.03% for sRNATargetNB, and 80.55%, 72.73%, and 80.65% for sRNATargetSVM, respectively. Therefore, the presented models provide support for experimental identification of sRNA targets. Following figure displays the relationship between the number of selected features and LOOCV classification accuracy or stability index. The stability index was the average of prediction accuracies from the minor parts in 1000 simulations with partition ratio 75%.

The relationship between the number of features and LOOCV classification accuracy or stability index