RNA has recently emerged as the ideal candidate as a building block to create the next generation of molecular medicines, with the potential to speci?cally disable or manipulate the genes involved in disease. RNA's functional versatility is exempli?ed by the development of myriad novel RNA-based synthetic biology tools, including virus-detecting RNA devices, CRISPR/Cas9 genome editing, and RNA silencing. The ability to design RNA elements is central to gaining precise control over these systems and customizing them for use as RNA-based medicines. However, doing so relies upon a thorough understanding of the energetics of RNA interactions and computational models has proven to be inadequate to quantitatively account for the biophysical behavior of RNA molecules. This lack of predictive models has been a major barrier to the development of life-saving therapies. Fortunately, modern high-throughput experimental techniques have enabled us to collect massive datasets of biophysical parameters, and recent innovations in deep learning have shown unprecedented power in the extraction of relevant features from such datasets. Thus, I propose to train and evaluate deep learning models for predicting the thermodynamic parameters governing RNA interactions. First, I will use convolutional neural network architectures to learn the energetic contributions of RNA motifs using high-throughput melting curve measurements on a speci?cally designed RNA library. Second, I will develop recurrent neural network models for predicting RNA-reporter af?nities to enable design of RNA-based sensors and diagnostics. As a result of these efforts, we will gain not only a nuanced understanding of the biophysics underlying RNA interactions but also quantitative models that will enable the design of RNA elements to control biological systems. This will give us the ability to take full advantage of RNA-based synthetic biology tools for RNA diagnostics and therapeutics.
The ability to design RNA interactions would enable precise control over RNA-based systems, including CRISPR/Cas9 genome editing and RNA silencing, that could become the next generation of therapeutics. Recent advances in high-throughput experimental approaches have enabled us to generate massive datasets of thermodynamic parameters covering RNA interactions, while algorithmic innovations in deep learning will enable us to leverage these datasets to gain novel biological insights. The proposed research would result in quantitative models that would enable us to take full advantage of RNA-based synthetic biology tools for RNA medicines.
