Optimizing nucleic acid sequences for a molecular data recorder

Abstract

We recently reported the design for a DNA nano-device that can record and store molecular signals. Here we present an evolutionary algorithm tailored to optimising nucleic acid sequences that predictively fold into our desired target structures. In our approach, a DNA device is first specified abstractly: the topology of the individual strands and their desired foldings into multi-strand complexes are described at the domain-level. Initially, this design is decomposed into a set of pairwise strand interactions. Then, we optimize candidate domains, such that the resulting sequences fold with high accuracy into desired target structures both (a) individually and (b) jointly, but also (c) to show high affinity for binding desired partners and simultaneously low affinity to bind with any undesired partner. As optimization heuristic we use a genetic algorithm that employs a linear combination of the above scores. Our algorithm was able to generate DNA sequences that satisfy all given criteria. Even though we cannot establish the theoretically achievable optima (as this would require exhaustive search), our solutions score 90% of an upper bound that ignores conflicting objectives. We envision that this approach can be generalized towards a broad class of toehold-mediated strand displacement systems.