Machine learning models to predict off-target effects in RNAi

---

Abstract

RNA interference (RNAi) has proved to be a valuable pathway process and successful tool for posttranscriptional gene regulation in prokaryotic and invertebrate systems. With high specificity, small interfering RNAs (siRNAs) induce mRNA degradation and thus inhibiting gene expression. This mechanism enables the design of synthetic siRNAs for use in mammalian cell loss-of-function experiments, drug target identification, and as a potentially viable therapeutic. However, the efficacy of RNAi in mammalian systems remains limited by target accessibility, positional preference, and off-target gene interactions. It has been shown that off-target effects are dominant in siRNA screens, mainly due to microRNA(miRNA)-like activity where each siRNA can target RISC complexes towards the 3’UTR of unintended targets and thus leading to off-target mRNA silencing. Experimental and computational approaches are being explored to understand the miRNA-like off target gene silencing mechanism in order to improve the specificity of RNAi in mammalian systems. Current miRNA target prediction algorithms present limited efficiency in identifying phenotypically-relevant transcript targets. The goal of this study is to train machine learning classifiers to recognize important descriptors of miRNA targeting and classify 3’UTR and siRNA features responsible for off-target mRNA silencing. Robust machine learning algorithms provide predictive models that suggest important descriptors involved in mRNA silencing mechanisms, which will allow mitigation of off-target effects in siRNA screens and improvement of the efficacy of engineered RNAi systems. Improved understanding of miRNA off-targeting will allow discovery of biologically-relevant pathway nodes and contribute in developing therapeutic solutions.

Mechanisms of RNA interference