*4.1.1. Use of artificial neural network for siRNA classification*

Artificial neural networks (ANNs), as they aim to mimic the working of biological networks through a connectionist approach to computation, provide a powerful method of identifying highly complex traits in data sets. ANNs are generally very efficient classifiers in case of complex patterns in the given data set as they can adaptively change their weighting parameters during the learning process. ANNs have been broadly applied in the biological sciences. The prediction quality and generalization capabilities of an ANN of fixed size depend on a sufficiently large training set of directly comparable data points.

Biopredsi siRNA designing algorithm from Novartis lab used Stuttgart Neural Net Simulator to train algorithms on a data set of 2182 randomly selected siRNAs targeted to 34 mRNA species [11]. It reliably predicted activity of 249 siRNAs of an independent test set (Pearson coefficient *r* = 0.66) and siRNAs targeting endogenous genes at mRNA and protein levels.

Computational Approaches for Designing Efficient and Specific siRNAs 269

siRNAs silence unintended transcripts in mainly two ways: transcripts with near perfect complementarity are cleaved while transcripts with imperfect complementarity are translationally repressed. mRNAs other than intended targets which exhibit near perfect sequence complementarity with the siRNA are likely to be degraded by the siRNA. This kind of off-targets can be avoided by choosing targets sites that do not have a large number

siRNAs down regulate a set of transcripts with 3' UTR complementarity to the 5' portion of the corresponding siRNA guide strand. These 5' ends of the guide strand resemble the seed region of endogenous microRNA and are responsible for target recognition. Such off-targets are regulated by translational repression like miRNA target regulation. This kind of off-

siRNAs can induce potential unwanted effects by activating innate immune system. Exogenous siRNAs are prone to be recognized by Toll-like receptors (TLRs), mainly TLR7, TLR8 and TLR9. TLR7 and TLR8 recognize synthetic siRNAs in a sequence dependent manner [16]. There seems to be preferential recognition of GU-rich sequences. AU rich sequences can also be immune stimulatory. Selecting siRNA sequences lacking GU rich regions can provide siRNAs with low immune stimulatory activity. Also presence of the motif "GUCCUUCAA" the 4-base motif "UGGC" in the siRNA is known to be immune stimulatory [17]. So, this motif should be avoided in the time of designing of siRNAs. The length of the siRNA is also an important factor for stimulation of immune response- the minimum length of siRNA to be recognized by innate immune system is in the range of 19

mRNAs other than intended targets which exhibit near perfect sequence complementarity with the siRNA are likely to be degraded by the siRNA. This kind of off-targets can be avoided by choosing targets sites that do not have many consecutive base homologies with any other mRNA. Actually siRNAs can potentially silence transcripts with more than 11 base complementarity including base matches corresponding its 9th-11th nucleotides. But as finding unique 11 base target site is impossible, the siRNA designing algorithms try to find unique target sites that do not have 15 or more consecutive base homology with other

siRNAs down regulate a set of transcripts with 3' UTR complementarity to the 5' portion of the corresponding siRNA guide strand. These 5' ends of the guide strand resemble the seed region of endogenous miRNA which is responsible for target recognition. Such off-targets are regulated by translational repression like miRNA target regulation. That is why this

target cannot be fully avoided but can be reduced by computational design.

of consecutive base homology with any other mRNA.

**6.1. Stimulation of innate immune response** 

**6.2. Near perfect complementarity with other mRNAs** 

nucleotides.

transcripts.

**6.3. miRNA-like off-target effect** 
