RNA secondary structure packages evaluated and improved by high-throughput experiments

Abstract

Despite the popularity of computer-aided study and design of RNA molecules, little is known about the accuracy of commonly used structure modeling packages in tasks sensitive to ensemble properties of RNA. Here, we demonstrate that the EternaBench dataset, a set of more than 20,000 synthetic RNA constructs designed on the RNA design platform Eterna, provides incisive discriminative power in evaluating current packages in ensemble-oriented structure prediction tasks. We find that CONTRAfold and RNAsoft, packages with parameters derived through statistical learning, achieve consistently higher accuracy than more widely used packages in their standard settings, which derive parameters primarily from thermodynamic experiments. We hypothesized that training a multitask model with the varied data types in EternaBench might improve inference on ensemble-based prediction tasks. Indeed, the resulting model, named EternaFold, demonstrated improved performance that generalizes to diverse external datasets including complete messenger RNAs, viral genomes probed in human cells and synthetic designs modeling mRNA vaccines.

Deep learning models for predicting RNA degradation via dual crowdsourcing

“…In addition to these experimental data, Kaggle participants were also provided with features related to RNA secondary structure computed from available biophysical models to use if they wished. These features included N × N base-pairing probability matrices from EternaFold 22 , a recently developed package with state-of-the-art performance on RNA structural ensembles; dot-parenthesis notated minimum free energy (MFE) RNA secondary structure from the ViennaRNA package 23 ; and a six-character featurization of the MFE structure calculated using bpRNA 24 .…”

“…Conversely, the model architectures presented here may also prove to have useful biological applications in predicting only SHAPE reactivity data. Future directions for model development include training such models on larger chemical mapping datasets from more diverse experimental sources 22 and integrating into inference frameworks for RNA structure prediction 22 , 33 .…”

“…As a starting point for Kaggle teams, we supplied a collection of features for each RNA sequence, including the minimum free energy (MFE) structure according to the ViennaRNA 2 energy model 23 , ‘loop type’, or secondary structure type assignments generated with bpRNA 24 (S = Stem, I = Internal loop, B = Bulge, H = Hairpin, M = Multiloop, X = external loop, E = end, terminology adopted from bpRNA) and the base-pair probability matrix according to the EternaFold 22 energy model. These features were generated using Arnie ( https://github.com/DasLab/arnie ).…”

AliNA – a deep learning program for RNA secondary structure prediction

“…Secondary structures of these sequences were obtained using the RNAfold software. This program was selected since comparison of several method showed advantage of this method [16a,33].…”

Leadership Through Citizen Science Games

“…For instance, in EteRNA , players need to help design ribonucleic acid (RNA) molecules to help create new medicines or even vaccines. At the beginning of the COVID‐19 pandemic, EteRNA 's developers posited a series of challenges that helped in developing mRNA vaccinations (Schrier, 2021b; Wayment‐Steele et al., 2022). In another game, Play to Cure: Genes in Space , players need to collect fuel in a space adventure game.…”