No abstract

Information technologies enable programmers and engineers to design and synthesize systems of startling complexity that nonetheless behave as intended. This mastery of complexity is made possible by a hierarchy of formal abstractions that span from high-level programming languages down to low-level implementation specifications, with rigorous connections between the levels. DNA nanotechnology presents us with a new molecular information technology whose potential has not yet been fully unlocked in this way. Developing an effective hierarchy of abstractions may be critical for increasing the complexity of programmable DNA systems. Here, we build on prior practice to provide a new formalization of ‘domain-level’ representations of DNA strand displacement systems that has a natural connection to nucleic acid biophysics while still being suitable for formal analysis. Enumeration of unimolecular and bimolecular reactions provides a semantics for programmable molecular interactions, with kinetics given by an approximate biophysical model. Reaction condensation provides a tractable simplification of the detailed reactions that respects overall kinetic properties. The applicability and accuracy of the model is evaluated across a wide range of engineered DNA strand displacement systems. Thus, our work can serve as an interface between lower-level DNA models that operate at the nucleotide sequence level, and high-level chemical reaction network models that operate at the level of interactions between abstract species.

Reversible chemistry allowing for assembly and disassembly of molecular entities is important for biological self-organization. Thus, ribozymes that support both cleavage and formation of phosphodiester bonds may have contributed to the emergence of functional diversity and increasing complexity of regulatory RNAs in early life. We have previously engineered a variant of the hairpin ribozyme that shows how ribozymes may have circularized or extended their own length by forming concatemers. Using the Vienna RNA package, we now optimized this hairpin ribozyme variant and selected four different RNA sequences that were expected to circularize more efficiently or form longer concatemers upon transcription. (Two-dimensional) PAGE analysis confirms that (i) all four selected ribozymes are catalytically active and (ii) high yields of cyclic species are obtained. AFM imaging in combination with RNA structure prediction enabled us to calculate the distributions of monomers and selfconcatenated dimers and trimers. Our results show that computationally optimized molecules do form reasonable amounts of trimers, which has not been observed for the original system so far, and we demonstrate that the combination of theoretical prediction, biochemical and physical analysis is a promising approach toward accurate prediction of ribozyme behavior and design of ribozymes with predefined functions.

T he overwhelming majority of small nucleolar RNAs (snoRNAs) fall into two clearly defined classes characterized by distinctive secondary structures and sequence motifs. A small group of diverse ncRNAs, however, shares the hallmarks of one or both classes of snoRNAs but differs substantially from the norm in some respects. Here, we compile the available information on these exceptional cases, conduct a thorough homology search throughout the available metazoan genomes, provide improved and expanded alignments, and investigate the evolutionary histories of these ncRNA families as well as their mutual relationships.

Machine learning (ML) and in particular deep learning techniques have gained popularity for predicting structures from biopolymer sequences. An interesting case is the prediction of RNA secondary structures, where well established biophysics based methods exist. The accuracy of these classical methods is limited due to lack of experimental parameters and certain simplifying assumptions and has seen little improvement over the last decade. This makes RNA folding an attractive target for machine learning and consequently several deep learning models have been proposed in recent years. However, for ML approaches to be competitive for de-novo structure prediction, the models must not just demonstrate good phenomenological fits, but be able to learn a (complex) biophysical model. In this contribution we discuss limitations of current approaches, in particular due to biases in the training data. Furthermore, we propose to study capabilities and limitations of ML models by first applying them on synthetic data (obtained from a simplified biophysical model) that can be generated in arbitrary amounts and where all biases can be controlled. We assume that a deep learning model that performs well on these synthetic, would also perform well on real data, and vice versa. We apply this idea by testing several ML models of varying complexity. Finally, we show that the best models are capable of capturing many, but not all, properties of RNA secondary structures. Most severely, the number of predicted base pairs scales quadratically with sequence length, even though a secondary structure can only accommodate a linear number of pairs.