RNA-deg-rates from the OpenVaccine project

What this is

Calibration data for models of RNA degradation.

Tips for modelers:

• Key data are in RNA_deg_rates.csv -- that's all you need.
• You'll want to sum per-nucleotide degradation rates for the sequence between start_pos and end_pos specified in RNA_deg_rates.csv and compare to the k_deg values there. The k_deg_err values there give estimates of experimental uncertainties.
• Example predictions -- we've made predictions for degradation rates for each nucleotide in the RNA sequences specified in all_RNA_sequences.csv with EternaFold in all_RNA_P_UNP_EternaFold.csv; these give rise to k_pred_eternafold values.
• To allow for experiment to experiment variation in temperature/pH and overall hydrolysis rate, may want to float a per-experiment scale factor that varies by up to ~2x. Index on the data_file column. (Authors of this repo haven't tried this yet.)

More info on the experiments:

• Overall hydrolysis rates (k_deg) for RNA molecules are in units of 1/hr.
• Several experimental methods used and a wide length range; see notes on data sources below.
• Data were acquired by Ann Kladwang and largely analyzed by R. Das and Do Soon Kim as part of the OpenVaccine initiative; samples and experimental design for PERSIST-seq data included important contributions from Barna lab.
• All measurements taken under accelerated degradation conditions 10 mM MgCl2, 50 mM Na-CHES, pH 10 in the Das laboratory at Stanford University. Note that Mg(2+) ions and high pH all serve to accelerate degradation; conditions may mimic environment of RNA in some cationic formulations. Conditions ensured experiments could be completed in 24 hrs; rates measured under other conditions (no Mg2+, lower pH) correlate well.
• Measurements acquired at ambient temperature of ~24 °C; note that due to unexpected laboratory temperature fluctuations, temperature may have varied between 21 to 25 °C, which may give rise to variations in overall modification rate between different days.

Contact Rhiju Das (rhiju [at] stanford.edu) with any questions.

Files

• RNA_deg_rates.csv: compilation of data, error estimates, RNA sequences, and start/end positions over which to sum per-nucleotide predictions, predictions from one model (EternaFold SUP; see below).
• prepare_data/: archive of data and prediction files, and MATLAB script rhiju_make_master_plot_script.m used to assemble the data file.

Example of model fit

Model plotted above is:

• EternaFold secondary structure ensemble (EternaFold is Contrafold-SE with parameters retrained based on data sets collected on Eterna; note no overlap of those training data with the degradation data here). See this paper.
• Predictions are summed unpaired probability (SUP), as described in this paper. Note that sum is not necessarily over entire RNA. For experimental readouts that are based on reverse transcription-PCR, the observed degradation rates reflect degradation over only the window that is encompassed by the primer pairs. (These positions are given here as start_pos and end_pos in the data files).
• For RNA's synthesized with pseudouridine (PSU) or N-1-methyl-pseudouridine (m1PSU), predictions were zeroed out at the linkage 3' to the subsitution site, reflecting observations noted in Leppek et al., 2021. (Reducing or zeroing the predicted degradation 5' to the substitution is worth checking too, just not done here.)
• The proportionality constant plotted is 1/3 of the one predicted by the model of Li and Breaker at 10 mM Mg(2+), 25 °C, 25 mM monovalents (7.91E-05/nt/hr).
• An alternative fit based on DegScore does not go through all the points very well; see analysis notebook in DegScore repo.

Data sources

The experiments have largely been described and archived elsewhere, and involve 5 different experimental methods.

• Eterna Roll your own structure data are based on the In-line-seq protocol on short RNAs (~100 nt), run on Illumina Next-Seq instruments, described in Fig. 3 of the Leppek et al., 2021. Single-nucleotide-resolution data are available on the RNA Mapping Database and were the public training and leaderboard data in the Kaggle Open Vaccine challenge, described in Wayment-Steele et al., 2021c. The data in this archive are the summed degradation rates over the nucleotides probed in In-line-seq, without any further normalization except to convert to rates (per hour); note that the RMDB data were normalized. These summed degradation rates correlate quite well with estimates based on 'drop out' of full-length RNA's from the In-line-seq experiment (unpub. analysis, R.D.)
• Eterna Roll your own structure 2 are similarly summed In-line-seq degradation rates provided private leaderboard data forthe Kaggle Open Vaccine challenge, described in Wayment-Steele et al., 2021c.
• P4P6-2HP control,Eterna Roll your own structure Followup,Eterna Roll your own structure Followup 5mC,Eterna Roll your own structure Followup PSU,Eterna Roll your own structure Followup m1PSU are based on in-line hydrolysis profiles read out through capillary electrophoresis assays (ABI 3130), using a protocol analogous to the one in Kladwang et al., 2014 though without any normalization.
• eGFP, MEV,NLuc Barna lab,NLuc Eterna are degradation rates for RNA molecules in the PERSIST-seq in solution stability measurements in Fig. 3 of [Leppek et al., 2021]; see also Table S1 in that paper. Lengths go up to ~1000 nts. The RNA's are actually mRNAs, but in these and other experiments were characterized without capping and poly(A) tails to enable easy comparison to CE measurements.
• full mRNA, gel degradation (older) are (previously) unpublished data for RNA degradation rates measured through conventional PAGE gels and ImageJ analysis.
• full mRNA, Bioanalyzer,full mRNA 5mC, Bioanalyzer,full mRNA PSU, Bioanalyzer are degradation rates based on running nanoluciferase RNA through capillary electrophoresis measurements on Agilent Bioanalyzer instruments; see last panels of Fig. 3 of Leppek et al., 2021. See also repo of analysis scripts.
• Nluc-combo18, Nluc-combo18-PSU,Nluc-final24-PSU are measurements on 'mix-and-match' combinations of CDS and UTR's described in Fig. 4 of Leppek et al., 2021. The last set of measurements ('final 24 nanoluciferase') are in Table S5 of the paper.
• S2P, m1PSU, S2P are based on bioanalyzer CE measurement on SARS-CoV-2 Spike protein mRNA's designed in Eterna's OpenVaccine Round 7 challenge. These long mRNAs (~4000 nts long) were challenging to synthesize and characterize by Bioanalyzer, hence the large error bars.