Computational methods for functional analysis of plant small RNAs using the RNA degradome

Small RNAs (sRNAs) are a broad class of short regulatory non-coding RNAs that play critical roles in many important biological pathways. They suppress the translation of messenger RNAs (mRNAs) by directing the RNA-induced silencing complex to their sequence-specific mRNA target(s). In plants, this typically results in mRNA cleavage and subsequent degradation of the mRNA. Cleaved mRNA frag­ments can be captured on a genome-wide scale using a high-throughput sequencing technique called degradome sequencing, which can then be used to identify causal sRNAs. Recent improvements to sequencing technologies have resulted in typical se­quencing experiments now producing millions of unique reads. This has led to new challenges in bioinformatics regarding the computation time and resources required to perform sRNA and degradome data analyses. In this thesis, we present three new sRNA and degradome analysis tools that we have developed called PAREsnip2, PAREameters and NATpare. PAREsnip2 is a tool we developed to predict sRNA targets, on a genome-wide scale, using degradome data and configurable targeting rules. Employing novel sequencing encoding and data structures, PAREsnip2 outperforms existing tools in computation time, at times by more than two orders of magnitude, with minimal computational resource requirements. PAREameters is a computational method for inference of plant microRNA targeting rules, using the degradome, that can then be employed by PAREsnip2. Benchmarking on multiple A. thaliana datasets show that the computationally inferred criteria outperform currently used criteria in terms of sensitivity on all datasets while maintaining precision on most. NATpare is a tool for high-throughput prediction and functional analysis of nat-siRNAs using the degradome. NATpare is the first tool of its kind to combine nat-siRNA prediction with functional analysis using the degradome. Compared to current methods, our new algorithm speeds up computation time by over two orders of magnitude when analysing an A thaliana dataset. We also demonstrate that it is the only computational method able to complete analyses of non-model organisms within a reasonable time frame. We exemplify the use of these computational methods by performing functional analysis of CMV D-satRNA derived sRNA in S. lycopersicum to better understand their role in virus induced plant death