RNA design rules from a massive open laboratory

Abstract

Self-assembling RNA molecules present compelling substrates for the rational interrogation and control of living systems. However, imperfect in silico models--even at the secondary structure level--hinder the design of new RNAs that function properly when synthesized. Here, we present a unique and potentially general approach to such empirical problems: the Massive Open Laboratory. The EteRNA project connects 37,000 enthusiasts to RNA design puzzles through an online interface. Uniquely, EteRNA participants not only manipulate simulated molecules but also control a remote experimental pipeline for high-throughput RNA synthesis and structure mapping. We show herein that the EteRNA community leveraged dozens of cycles of continuous wet laboratory feedback to learn strategies for solving in vitro RNA design problems on which automated methods fail. The top strategies--including several previously unrecognized negative design rules--were distilled by machine learning into an algorithm, EteRNABot. Over a rigorous 1-y testing phase, both the EteRNA community and EteRNABot significantly outperformed prior algorithms in a dozen RNA secondary structure design tests, including the creation of dendrimer-like structures and scaffolds for small molecule sensors. These results show that an online community can carry out large-scale experiments, hypothesis generation, and algorithm design to create practical advances in empirical science.

Keywords: RNA folding; citizen science; crowdsourcing; high-throughput experiments.

Fig. 1.

EteRNA workflow. Each week, participants (A) design sequences that can fold into a target RNA structure in the sequence design interface and (B) review and vote for the best designs with the voting interface. (C) At the end of the round, the eight top-voted sequences are synthesized and verified by single nucleotide resolution chemical reactivity measurements. (D) The experimental results are published online and available for review in the results viewer. Participants then create new hypotheses and (A) start the next experimental cycle or (E) submit design rules learned from the results (text) that are codified and automatically ranked based on scores obtained to date (numbers).

Fig. 2.

Phase I puzzles and results in order of puzzle posting date. Top shows a target structure and title for each puzzle (A–F). Nucleotide coloring in target structures indicates the ideal SHAPE reactivity (gold for high reactivity and blue for low reactivity). Middle gives the single nucleotide resolution reactivity data measured for all designs. Yellow stripes indicate bases that should show high reactivity if the target secondary structure (Top) is formed. Bottom shows a summary of structure mapping scores for designs from the RNAInverse (black) and NUPACK (gray) algorithms compared with EteRNA participants (colored symbols; ordered by score within each design round). Each design was subjected to SHAPE chemical reactivity mapping in two solution conditions, 1 M NaCl (circles) and 10 mM MgCl₂ (squares), with 50 mM Na-Hepes (pH 8.0) at 24 °C. The colored border lines connect designs within the same round.

Fig. 3.

RNA design rules proposed by EteRNA participants. (A) The best designs from each design agent (EteRNA participants, NUPACK, and RNAInverse) for the last target shape of phase I (Fig. 2F); the nucleotide coloring gives experimental chemical reactivity and is identical to the coloring used in Fig. 2. The designs are annotated with violations of the top 5 rules of 40 rules proposed by participants, which were assessed by sparse linear regression. (B) The five rule statements used for EteRNABot. The numerical parameters in brackets were optimized to best explain the results from a training set based on starting values proposed by participants (Materials and Methods).

Fig. 4.

Phase II puzzles and results, including the EteRNABot algorithm, in order of puzzle posting date. (A–E) Multijunction target structures distinct from each other and the phase I structures in topology. (F–I) Puzzles giving the binding site for FMN as a fixed sequence that must be displayed as an internal loop (boxed in the target structure; J shows the sequence) within a complex structure. The coloring scheme is identical to the coloring scheme used in Fig. 2, with EteRNABot shown in magenta. (K–N) FMN association constants for all designs synthesized for the last four puzzles measured by dimethyl sulfate chemical mapping as a function of FMN concentration. Measurements on a simple construct displaying FMN binding (green dashed line) give the best possible association constant, which is achievable only with correct secondary structure folding. The down arrows in K–N mark RNAInverse designs that gave no observable FMN binding, setting the upper bounds on the association constants.