miRNA prediction software
The clean data of small RNA sequencing (miRNA and siRNA reads) were mapped to G. barbadense using Bowtie, which allowed 200 multiple mapping positions and zero mismatch for each read. We adopted structure-based annotation and probability-based annotation to predict miRNA loci as suggested by Paterson et al. [22]. For the structure-based annotation, RNAfold was employed to predict secondary structures and miRcheck was used to evaluate secondary structures [65]. We then utilized miRDP to filter the putative precursors of the structure-based annotation [66]. The cutoff value of the largest miRNA family size was set at 30 owing to the genome doubling of teraploid relative to diploid. All the annotated mature miRNAs were searched against the miRBase (Release 20) to categorize into cotton conserved and non-conserved miRNA gene families [67]. We also employed the CleaveLand pipeline to predict putative miRNA targets based on the degradation data [68]. The bona fide miRNA targets were detected based on the criterion suggested by Addo-Quaye et al. [69]. After eliminating miRNA reads from small sequencing datasets, the remaining were then regarded as siRNAs. Wang et al.,2015 NP
Target predict
Annotated mRNA database of upland cotton from NCBI and assembled contigs from our cotton EST database was employed to identify miRNA targets by Target-align with default
parameters(Xie and Zhang, 2010). To validate predicted miRNA targets, Cleveland (Version 4.3, http://axtell-lab-psu.weebly.com/cleaveland.html) (Addo-Quaye et al., 2009) was used
to do degradome analysis with some modifications, where maximal mismatches between miRNAs and target sites are less than four and no mismatches is allowed on the 10th and
11th nucleotides on miRNAs (Meyers et al., 2008). Xie et al.,2015 PBJ