Designing Molecular RNA Switches with Restricted Boltzmann Machines

Riboswitches are structured allosteric RNA molecules that change conformation in response to a metabolite binding event, eventually triggering a regulatory response. Computational modelling of the structure of these molecules is complicated by a complex network of tertiary contacts, stabilized by the presence of their cognate metabolite. In this work, we focus on the aptamer domain of SAM-I riboswitches and show that Restricted Boltzmann machines (RBM), an unsupervised machine learning architecture, can capture intricate sequence dependencies induced by secondary and tertiary structure, as well as a switching mechanism between open and closed conformations. The RBM model is then used for the design of artificial allosteric SAM-I aptamers. To experimentally validate the functionality of the designed sequences, we resort to chemical probing (SHAPE-MaP), and develop a tailored analysis pipeline adequate for high-throughput tests of diverse homologous sequences. We probed a total of 476 RBM designed sequences in two experiments, showing between 20% and 40% divergence from any natural sequence, obtaining ˜ 30% success rate of correctly structured aptamers that undergo a structural switch in response to SAM. Overall design: DNA oligonucleotides representing the 206 SAM-I aptamer natural sequences, and the two batches (100 and 450) of artificial sequences, preceded by the T7 promoter (5'CGGCGAATCTAATACGACTCACTATAGG3') and followed by a tag sequence representing a 10 nucleotide bar code unique for each aptamer and a primer binding site, were purchased as an oligonucleotide pool (Twist bioscience®). The Tag sequence was designed to avoid interference with the aptamer secondary structure using RNAFold. The oligo pool was PCR amplified using the T7 promoter as forward primer and and five different reverse primers. RNAs were in vitro transcribed by using the T7 RNA polymerase and were checked for the absence of aberrant products on a 1% agarose gel. SHAPE chemical probing was performed as described previously [1]. Then, RNAs were then reverse transcribed with the Superscript III reverse transcriptase (Invitrogen®) and NGS libraries were prepared using NEBNext Ultra II DNA Library Prep Kit (New England Biolabs®). Final products were sequenced by using the illumina technology (NextSeq 500/500 Mid 2x150 flowcell). Sequencing data were analyzed and reactivity maps were derived using ShapeMapper2. [1] N. A. Siegfried, S. Busan, G. M. Rice, J. A. Nelson, and, K. M. Weeks. Rna motif discovery by shape and mutational profiling (shape-map). Nature methods, 11(9):959–965, 2014.