Sensitive and Accurate Quantification Of Small RNAs At Single-nucleotide Resolution From Deep Sequencing Data - This project, entitled “Sensitive and Accurate Quantification Of Small RNAs from Deep Sequencing Data At Single-nucleotide Resolution,” focuses on the development, software implementation, and benchmarking of statistical methods that can correctly quantify small non-coding RNAs (sncRNAs) in next-generation se- quencing (NGS) data while adhering to constraints imposed by recent experimental findings. NGS revolutionized sncRNA studies and led to the discovery of previously unknown molecular classes, such as the microRNA isoforms (isomiRs), transfer RNA-derived fragments (tRFs), and others. The available tools for NGS sncRNA data analyses use invalid assumptions inherited from methods designed for messenger RNA data. Recent findings in the sncRNA field have created a critical need for new methods that correctly estimate sncRNA abundance from NGS data. Indeed, based on these findings, we now know that: (1) the typical precursor transcript (e.g., miRNA, tRNAs) produces simultaneously multiple mature “sibling sncRNAs” (e.g., isomiRs, tRFs), the result of heterogeneous enzymatic cleavage, RNA editing, post-transcriptional non-templated nucle- otide additions, polymorphisms, or somatic mutations; (2) some – but not all – sibling sncRNAs that differ by as few as 1-2 nucleotides exhibit distinct abundance patterns that depend on the sncRNA’s sequence, context (tissue or disease type), and personal attributes (sex, ancestry, age). Consequently, methods that estimate abundance by lumping together all siblings are incorrect, and so are methods that treat each sibling as an independent true sncRNA. Any method that aims to correctly quantify sncRNA abundances must have two properties. First, it should distinguish abundant “artifact variants” from abundant real sncRNAs without loss of sensitivity: these variants are common and introduced during the reverse transcription step, polymerase chain reaction step, or during the sequencing cycles. Second, it should split sib- ling sncRNAs into two groups, correlated and uncorrelated ones: not doing so compromises statistical power and downstream interpretability and conclusions. We are unaware of any methods that exhibit these properties. In Aim 1, we will develop a computational method for denoising sncRNA NGS data based on Bayesian statistics that infers the error rate distribution from the data being analyzed and labels each NGS read as a “true” or “fake” sncRNA. In Aim 2, we will develop a method for decorrelating sncRNA profiles that groups sibling sncRNAs with highly correlated abundances into a single “meta-sncRNA”. We will thoroughly benchmark both methods using simulated data and public datasets from NIH’s Sequence Read Archive. We will implement both methods as fast, cross-platform, user-friendly tools that can be used by biologists with minimal computer skills. A successful completion of the project will close a major gap and create unique research opportunities. One such example of an important application is to causally link regulatory sncRNAs with protein-coding genes and complex human traits by integrating denoised and decorrelated sncRNA profiles with matched genotype data.
