Rational Design of Ribozyme Switches through Big Data and Machine Learning

Calvin M. Schmidt, Christina D. Smolke

Stanford University; Stanford, California

RNA switches are a class of functional RNAs that support conditional regulation of gene expression. These genetic switches have broad application in the design of biological systems responsive to changes in their environment with implications for the fields of metabolic engineering, environmental sensing, and therapeutics. The mechanisms underlying the activity of RNA switches rely on programmed structural conformational changes encoded in the sequence of the RNA molecule, implying that an improved elucidation of relationships between sequence, structure, and function can be used to improve our capacities for de novo design. One type of RNA switch is the ribozyme switch, in which binding of the target ligand changes the structure and thus cleavage activity of the switch. Here, we leverage sequencing data and machine learning methods to relate the sequence and structure of RNA molecules to their function and develop models that predict the function of novel ribozyme switches. Through prior work, we have access to data on the activity of hundreds of thousands of ribozyme switch sequences. Using automated structural analysis and machine learning methods, we leveraged a wealth of sequence-activity data on hundreds of thousands of ribozyme switches to develop mathematical models that predict the in vivo activity of a given ribozyme based on its sequence. We have used these models for the de novo design of novel ribozyme switches that exhibit changes in in vivo gene-regulatory activity upon introduction of a target ligand. Our work demonstrates that computational tools can be used to convert RNA sequences into quantitative features that represent key sequence and structural motifs and that these features can be used to develop models that can learn how interactions between these motifs translates to the activity of the functional RNA molecule. The predictive models we developed provide a broadly accessible tool for designing custom biosensors for use in engineering novel biological systems