Sirna target finder and design tool

The increased potency of these reagents is thought to relate to linkage between Dicer processing and RISC loading [7]. Increased antisense loading will result in increased mRNA cleavage. Figure 2. Design of a DsiRNA. The ability of an siRNA to silence gene expression is predominantly determined by its sequence, and not all target sites are equal [5, 8].

In addition to the actual sequence, other considerations, such as cross-hybridization and chemical modifications, can alter the effectiveness of the siRNA [5]. Like targeted effects, off-target effects OTEs are dose dependent. Therefore, it is important to establish dose-response profiles for all siRNAs and always use the lowest concentration of siRNA that will provide adequate target knockdown.

An additional measure to prevent OTE bias is to ensure that at least two, and ideally three, independent siRNAs against a target give the same result [5]. Use of siRNAs in vivo shows great potential as both research tools and as therapeutic agents [11]. For more information on the status of RNAi in therapeutics, see the recent review by Vaishnaw et al.

Before you begin RNAi studies in vivo , consider the following issues: site selection, compound design and chemistry, controls, route of administration, and use of a delivery vehicle [11]. To find the best candidates, it is very important to validate siRNA duplexes in vitro before moving on to in vivo experiments. In addition, choose more than one effective siRNA for each target to be tested in order to rule out false positive results caused by off-target effects [11]. Intuitively, for each loop, if the segment enclosed by the enclosing branches with respect to the target site is small relative to the total length of the loop, the target site is more difficult to be accessed.

Therefore, we measure the ratio between the length of the segment enclosed by the enclosing branches with respect to the target site and the total length of the loop. Whether a repelling loop is considered to be big, we measure the length of the loop.

The thresholds for repelling loops and big repelling loops are obtained by using AI techniques, which we will describe in the next section. The filtering algorithm Based on the concept of repelling loops and big repelling loops, we have devised a filtering algorithm to filter potential ineffective siRNAs. The details are as follows. The MFOLD algorithm usually reports over 10 secondary structures, each with a free energy indicates the stability of the corresponding structure. We focus on the five structures that have the lowest free energy, i.

The filtering condition is as follows:. The filtering algorithm checks the filtering condition for each candidate siRNA and report those that are not filtered. In addition to the repelling loops, big repelling loops we mentioned in the previous section, we have also considered the following two factors that are related to our observations.

The number of branches in a repelling loop : Intuitively, if the number of branches increases, branches will be closer together, so if a target site is enclosed by the branches, it may be difficult to access it. The free bases in the target site : Free base has higher free energy and may interfere siRNA activity. Hence, we also consider the number of free bases in the target site. The total weight of the free bases will be used in the decision tree training.

We repeatedly train the decision tree by fixing the values of r in the range 0. For a particular value of r , we compute the following attributes for each siRNA. Recall that when considering the secondary structures, we use the five most stable structures reported by the MFOLD algorithm. Among these structures, select three that have more repelling loops and we break the ties by selecting the one with the lower free energy i.

The average number of branches of all the repelling loops in the three structures selected in Step 1. Note that the two branches that enclose the target site are not counted. The average of the total weight of free bases in the three structures selected in Step 1. We then train the decision trees by including different attributes as follows.

As a result, we have four combinations: including both the unpaired weight and the number of branches, including either one of them, and including none of them. Results Four decision trees are returned for each experiment. For each decision tree, we compute the rate of correct classification for ineffective and effective siRNAs. Figure 2 shows the classification rate of the decision trees for all the experiments. In each experiment, we only report the best decision tree, i.

The classification rate for ineffective and effective siRNAs is and In fact, when the repelling loop threshold is 0. We also make use of the support vector machine Joachims, to see if the classification is consistent with that of the decision tree. The support vector machine obtained has a similar performance as the decision tree we obtained.

Precisely, the classification rate for both ineffective and effective siRNAs are the same as that of the decision tree, and even further, the sets of siRNAs that are classified as ineffective and effective are the same for both the decision tree and the support vector machine.

These results show that the attributes and the thresholds are selected appropriately. As a remark, we have tried to use the rules to select siRNAs directly and compared its performance with the random selector. The results show that the rules perform better than the random selector.

In this paper, we have proposed a scheme to evaluate existing siRNA design tools based on the published effective and ineffective siRNAs.

In the scheme, the output of each design tool is compared with a set of randomly selected siRNA candidates. The results show that existing tools are not good at filtering ineffective siRNAs. We also propose a filtering algorithm to filter potential ineffective siRNA candidates from the output of existing tools.

The algorithm is based on two observations, namely repelling loops and big repelling loops, on secondary structures of the target mRNA. The rule for classifying potential ineffective siRNAs from other candidates is generated with the help of AI techniques, in particular, the decision tree and support vector machine.

The filtering algorithm is shown to be effective. The results of this paper provide evidence that the secondary structures should be considered for the design of siRNA. We are in the process of designing laboratory experiments to further verify our observations on secondary structures. Comparison of the net percentages against ineffective siRNAs before and after applying the filtering algorithm.

Comparison of the net percentages against effective siRNAs before and after applying the filtering algorithm. Ambion, S. Caplen, N.

Natl Acad. USA 98 — Cui, W. Dharmacon, M. Elbashir, S. Nature — Genes Dev. Fire, A. Holen, T. Nucleic Acids Res. Jacque, J. Joachims, T. In Scholkopf, B. Lin, J. First: The sequence-based rules for defining efficient siRNAs in the previous si-Fi version was replaced by a combination of thermodynamic calculation and sequence-based rules, resulting in better prediction accuracy. Second: An additional calculation for the local accessibility of the target sites is included in the new version.

To clearly distinguish the latest version from all previous releases, the name of the software was changed to si-Fi The results on HvMlo -related RNAi constructs, generated by previous si-Fi and the latest si-Fi21 , are compared in Table 2 , together with the experimentally derived mean silencing effect increased susceptibility to powdery mildew.

The results indicate that the novel version si-Fi21 provides higher sensitivity in HS-mode off-target prediction and better selectivity in HE-mode RNAi design than the previous version. The best correlation to the experimentally derived mean silencing effect is achieved by the si-Fi21 in HE-mode. Table 2 Summary of the prediction results of the previous version of si-Fi and the new si-Fi21 in bolt.

Using the selected HvMlo reference sequence, we compared the reliability of the prediction of selected softwares: siDirect Naito et al.

Since none of the online tools include the option to use a barley transcript database or a possibility to add a custom database, a direct comparison with the experimental results was not possible. Instead, a comparison was performed based on the number of recommended siRNAs predicted by the tools except siDirect. The siDirect is the only one of the selected online tools able to design long double-stranded RNAi. The design of the experiment is based on the expectation that the amount of efficient siRNAs is positively correlated with a high RNAi activity.

This is in agreement with the presented example of the HvMlo gene, where the —W construct Figure 4 is triggering the strongest RNAi effect. The HS-mode of si-Fi21 off-target search as well as the tested online tools are in less agreement to the experimental results. All RSI values are normalized to the empty vector control for the corresponding experiment and log 2 transformed. Lower values indicate stronger effect decreased susceptibility.

In Xu et al. The predictions made by the si-Fi21 tool on the respective sequences used in this study confirm the experimentally validated off-target effects Table 3. All genes with experimentally validated reduced transcript levels were successfully predicted by the HS-mode of si-Fi21 as off-targets.

A gene without predicted hits was also not affected on its transcript level by the RNAi transgene. RNAi has become an important research tool for studying gene function.

In contrast to other techniques that target the genomic DNA e. This can be, e. The knockdown nature of RNAi also allows targeting genes, which cause lethality after knockout.

Therefore, further development of tools and methods for RNAi applications remains an important task. The si-Fi21 software tool that we present here is specifically intended for design of long double-stranded RNAi that is widely used in non-vertebrate systems.

The prediction accuracy of si-Fi21 was validated by performing multiple experiments for estimation of the gene silencing effect of specifically designed RNAi triggers. The experimental outcome was validating the results to the predicted features. Since direct measurement of the decrease of the target gene transcript is technically not feasible in this model system, we have used the phenotypic effect of the silencing of the Mlo gene as a proxy for the silencing of its transcript Douchkov et al.

Interestingly, the 3MY construct shows the strongest effect. However, the differences between the constructs 0MY to 3MY are not statistically significant. As the used system is based on the interaction of two living organisms, it is associated with considerable variance. This drawback can be overcome with a high number of feasible independent repetitions Supplementary Table S2.

These ones have a higher chance to cause a silencing effect. Combination of both selection criteria is increasing the prediction power for RNAi efficiency. However, if the criteria for selection for siRNA efficiency are set too stringent, this may cause a non-detection of some potential off-targets.

Therefore, we have designed two different modes of siRNA selection. The high-sensitivity HS mode is using less stringent selection parameters only strand selection rules , and it is primarily designed for finding a maximum number of potential off-targets.

The high-efficiency HE mode is using stringent parameters stricter strand selection rules plus target site accessibility calculations , and it will select for siRNA with strong silencing potential. The number of predicted siRNA hits in HS- and HE-modes, related to the silencing effect of the corresponding constructs, provides an estimate of si-Fi21 prediction power Figure 3. The number of matching siRNA is progressively decreasing parallel to the decreasing overall identity of the RNAi trigger to the target.

These predictions are in good agreement with the observed Mlo silencing effects, which were significantly weaker, compared with the 0MY reference, from 4MY onward.

The results shown in Figure 3 suggest a putative threshold mechanism, where the silencing effect is exhibited only after reaching a certain siRNA pressure on the target. Identical experiments were performed with the bp window constructs. Although all constructs are expected to generate nearly the same number of matching siRNA 80 to 81 , their effect on susceptibility to Bgh differed substantially Figure 4. Only the —W construct had a significant effect on RSI.

While two out of the four inefficient constructs, —W and —W, were associated with relatively high numbers of HE-mode siRNA predictions, they were still below the number of HE-hits of the most active —W construct, which points into the direction of a threshold-dependent process of RNAi efficiency.

This threshold phenomenon might be related to the short time window for observing TIGS-induced phenotypes: A suboptimal construct might not be capable of silencing the target rapidly enough to reveal the phenotypic effect, although the same construct might silence the target gradually during its continuous exposure to the siRNAs e. All bp RNAi constructs gave rise to comparable numbers of HS-mode siRNAs indicating that this criterion is not sufficient to predict efficient silencing.

Taken together, the results presented here suggest that the maximum number of HE-mode siRNA molecules might be a useful indicator for optimal RNAi-construct design.

By combining the powerful BOWTIE-based sequence similarity search for putative siRNA targets together with the probability calculation of local target-site accessibility, and thermodynamics- as well as sequence-based prediction for strand selection, we have generated a Python-based software tool named si-Fi The software provides two different modeling modes—HS high sensitivity for off-target search and HE high efficiency for RNAi-construct design.

Each of the two modes is based on an own pipeline with adapted parameters Figure 5. The software offers the possibility to create custom sequence databases, allows flexible settings of parameters, and provides easy-to-interpret graphical and tabular outputs Figure 6. Figure 5 si-Fi21 workflow diagram.

The process begins with selection of the search mode. The off-target prediction is using the high-sensitivity HS mode settings for finding as many as possible putative off-targets as possible by minimizing the false-positive signals. Figure 6 si-Fi21 screenshots and example results of the RNAi design mode. The blue line indicates the total number of siRNA hits; the red line corresponds to the number of siRNA hits that match the selected criteria for efficiency; the red zone indicates the regions that may cause silencing of genes that were not selected as a target; in the green zone are the regions that have no match to other genes besides the selected target.

The datasets generated for this study can be found in e! SL performed research RNAi construct preparation and testing and programmed the software. TK performed research RNAi construct preparation and testing. MS contributed in writing and editing the manuscript. PS designed the research and contributed in writing the manuscript. DD designed the research and the software algorithms and performed research RNAi construct design.

MK and DD wrote the manuscript. All authors read and approved the final manuscript. The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. We express our appreciation to Dr. We thank Dr. In total 15 independent experiments were conducted, where each of the construct was tested in at least 5 of them.

The values indicate the Relatively Susceptibility Index RSI of each construct calculated as follows: Each experiment includes three empty vector controls. Acevedo-Garcia, J. Magical mystery tour: MLO proteins in plant immunity and beyond. LightRun Plate. LightRun Barcodes. SupremeRun Tube. SupremeRun Plate. SupremeRun Barcodes. Sequencing Projects. Primer Walking Service. Special Plate Sequencing. Re-Sequencing Projects.

GLP Sequencing. Additional Services. Sequencing Primers. Sequencing Accessories. Sample Shipment. Free Sample Pick-Up. Overnight Sequencing. NGS prepaid solutions. NGSelect Amplicons. NGSelect Ready2Load. Customised Solutions. Genome Sequencing. Transcriptome Sequencing. Sample Preparation. Bioinformatic Solutions.

Oncology solutions.