Modeling the influence of translation-elongation kinetics on protein structure and function

点击翻译按钮获取中文摘要

Project Summary mRNA degradation is an essential process in post-translational gene regulation, and influences protein expression levels in cells. In S. cerevisea the lifetime of mRNA ranges from 43 sec to 39 min, with a median half-life of 3.6 min. The molecular factors governing these differential degradation rates has long been an area of active research. Recently though, clear evidence has emerged that the codon optimality correlates with half- lives. At a mechanistic level, the emerging perspective is that some transcripts are translated quickly, and some slowly, and that transcripts in which ribosomes end up forming queues, much like a traffic jam of cars on a highway, are recognized by ubiquitin ligases such as Hel2 that trigger the RQC pathway to promote mRNA degradation. There are two major gaps in this field. The first is the capability to predict mRNA half-lives accurately from mRNA sequence features. The second is understanding at the molecular level how the distribution of codon translation speeds along a transcript’s coding sequence promote ribosome queues and hence degradation. In this proposal, a graduate student will combine the PI’s labs expertise in modeling the kinetics of translation and ribosome traffic with interpretable machine learning techniques to address these two gaps. In achieving this, the field will be advanced by having both predictive and explanatory models for how translation speed and codon usage differentially impacts the degradation rates of different mRNAs. Specifically, our first aim is to build an interpretable machine learning model to identify robust and predictive features governing mRNA degradation. Our second aim is to explain at the molecular level why these features influence degradation rates. We will do this in two ways. First, we will use the essential and predictive features resulting from the interpretable machine learning model to identify potential underlying mechanisms contributing to degradation. Second, we will simulate the movement of ribosomes on each transcript based on reported initiation and elongation rates to detect ribosome queues and provide an explanation for differential degradation rates. Finally, our third aim is to test the predictions coming from the models. For example, do the models from Aim 1 accurately predict mRNA half-lives when synonymous mutations are introduced? There is sufficient published data on transcriptome-wide mRNA half-lives on S. cerevisiae to train and test the machine learning models in Aim 1. Further, we have arranged for a machine learning expert to co-advise the graduate student on the second aim. This co-advisor is already a collaborator of the PI on other machine learning projects. Finally, a collaborator who has measured mRNA half-lives will further advise the student on the third aim. In summary, this training supplement will address cutting edge questions in the molecular biology and biophysics of mRNA lifetimes and provide the student the opportunity to get advanced training and expertise in machine learning, molecular modeling, and experimental techniques.