​​​Talk from Jorge Fernandez-de-Cossio-Diaz – Université Paris-Saclay, CNRS, CEA, Institut de Physique Théorique (IPhT)
​

Short abstract:​

​​Riboswitches are structured allosteric RNA that undergo a conformational change​ upon binding specific metabolites, eventually triggering a regulatory response. Their complex tertiary structure, ligand binding specificity, and ability to switch between competing folds, pose challenges for modeling and to the design of novel artificial riboswitches. I will describe a data-driven approach, whereby a model is trained on data of homologous molecules to extract sequence patterns relevant to their common functionality. Specifically, we focus on SAM-I riboswitch aptamers, and employ an interpretable two-layer generative model known as Restricted Boltzmann machines (RBM). RBM-generated sequences correctly capture the conservation, covariation and diversity of natural sequences. The model is then used to design new SAM-I riboswitch aptamers. To experimentally validate the switching functionality of the designed molecules, we resort to a high-throughput chemical probing technique known as SHAPE-MaP, and develop a tailored probabilistic analysis pipeline adequate for processing numerous diverse sequences. We probe a total of 476 RBM-designed and 201 natural aptamers. Designed molecules with high RBM scores, showing between 20% and 40% divergence from any natural homologue, display ~30% success rate of correctly binding and undergoing a structural switch in response to SAM, while the tested natural sequences have ~60% success rate. The capability of the designed molecules to switch conformation is related to fine energetic features of their structural components.

​