DesiRNA: structure-based design of RNA sequences with a replica exchange Monte Carlo approach

Abstract

Designing RNA sequences that form a specific structure remains a challenge. Current computational methods often struggle with the complexity of RNA structures, especially when considering pseudoknots or restrictions related to RNA function. We developed DesiRNA, a computational tool for the design of RNA sequences based on the Replica Exchange Monte Carlo approach. It finds sequences that minimize a multiobjective scoring function, fulfill user-defined constraints and minimize the violation of restraints. DesiRNA handles pseudoknots, designs RNA-RNA complexes and sequences with alternative structures, prevents oligomerization of monomers, prevents folding into undesired structures and allows users to specify nucleotide composition preferences. In benchmarking tests, DesiRNA with a default simple scoring function solved all 100 puzzles in the Eterna100 benchmark within 24 h, outperforming all existing RNA design programs. With its ability to address complex RNA design challenges, DesiRNA holds promise for a range of applications in RNA research and therapeutic development.

Graphical Abstract

Figure 1.

DesiRNA performance on the EteRNA100 benchmark dataset. (A) The EteRNA100 V1 dataset versus time points (1 min, 1 h and 24 h). Ninety of these puzzles were solved within 1 min, puzzles 64, 80, 85, 87 and 96 were solved within 1 h, and the remaining puzzles were solved within 24 h. (B) DesiRNA solved RNA puzzles (97 and 100) that were not solved within 24 h by any method to date.

Figure 2.

glmS ribozyme sequences design and experimental validation. (A) glmS design constraints were based on the RF00234 family consensus sequence, and we utilized the crystal structure of T. tengcongensis glmS ribozyme (PDB ID: 2Z75) as a reference. (B) glmS_D1 sequence designed with DesiRNA. (C) Cis-cleavage activity, IVT in the absence and presence of 2 mM GlN6P. The denaturing gel analysis of the IVT products shows the transcription of shorter RNA products in the presence of GlN6P. (D) In trans-cleavage activity, Cy5-labelled substrates were used along with the catalytic part of the respective designs and glmS_WT RNAs. Reaction products were collected at time points 0, 5, 15, and 30 min, and the samples were analyzed on a denaturing gel. The substrate (Pre, black) and product (Clv, red) bands are indicated with asterisks.

Figure 3.

Mom19-I and Mom19-II sequence design. (A) Mom19-I and Mom19-II sequence design constraints. (B) Sequence and secondary structure of the designed Mom19-I and Mom19-II (C) 3D representative of the prevalent structure observed in SimRNA folding simulations for Mom19-I. (D) 3D representative of the prevalent structure observed in SimRNA folding simulations for Mom19-II. (E) Crystal structure of RNA-II structure (RCSB PDB ID: 6IA2). The colored residues represent the bipartite 5′-GAD(N)₂HC-3′ motif.

Figure 4.

DesiRNA-designed RNA sequence with a comparable propensity to fold into two very different secondary structures. (A) One sequence and two secondary structures of the designed RNA: a helical stem with a double loop and two separate hairpins. (B) 3D models representing the largest five clusters generated by SimRNA simulations for the top designed solution. First and third clusters exhibit the helical stem conformation, while the second, fourth and fifth clusters exhibit the two-hairpin conformation.