Deciphering the rules of mRNA structure differentiation in vivo and in vitro with deep neural networks in Saccharomyces cerevisiae

Abstract

The structure of mRNA in vivo is influenced by various factors involved in the translation process, resulting in significant differentiation of mRNA structure from that in vitro. Because multiple factors cause the differentiation of in vivo and in vitro mRNA structures, it was difficult to perform a more accurate analysis of mRNA structures in previous studies. In this study, we have proposed a novel application of a deep neural network (DNN) model to predict the structural stability of mRNA in vivo by fitting six quantifiable features that may affect mRNA folding: ribosome density, minimum folding free energy, GC content, mRNA abundance, ribosomal initial density and position of mRNA structure. Simulated mutations of the mRNA structure were designed and then fed into the trained DNN model to compute their structural stability. We found unique effects of these six features on mRNA structural stability in vivo. Strikingly, the ribosome density of the structural region is the most important factor affecting the structural stability of mRNA in vivo, and the strength of the mRNA structure in vitro should have a relatively small effect on its structural stability in vivo. The recruitment of DNNs provides a new paradigm to decipher the differentiation of mRNA structure in vivo and in vitro. This improved knowledge on the mechanisms of factors influencing mRNA structural stability will facilitate the design and functional analysis of mRNA structure in vivo.