Secondary Structure Dataset
This is a compiled dataset of chemical mapping from RMDB. All data used from RMDB can be found in compiled_RMDB.tar.gz. Inside, the data is stored as parquet files (brotli compression). The format of the filenames are RMDB_dataset_<suffix>.parquet. The txt files contain a list of rdat filenames that make up the chemical mapping data of the parquet. RMDB_dataset_<integer>.parquet contains the data from only the ribonanza train and test set. RMDB_dataset_extra.parquet contains all the other RMDB data not used in the train and test set. For more granular detail the RMDB_dataset_extra was further split into extra_clean, extra_cotrans, extra_degradation, extra_invivo.
RMDB_dataset_extra_clean.parquet: normal chemical mapping data \RMDB_dataset_extra_cotrans.parquet: the RNA was folded cotranscriptionally (not refolded) \RMDB_dataset_extra_degradation.parquet: the reactivity measurements corresponds to where the RNA is degraded NOT reactivity. Some datapoints corresponds to different conditions (e.g. different pH) \RMDB_dataset_extra_invivo.parquet: the chemical mapping experiment was done in vivo
The parquet files have the following columns below.
| Column Name | Detail |
|---|---|
| seqID | Unique ID for the data. |
| sequence | RNA sequence. |
| modifer | Chemical modifer used to modify the RNA (mostly DMS and 2A3). |
| SNR | Signal-to-noise ratio. |
| reads | Total number of reads from the chemical mapping experiment. |
| temperature | Temperature of experiment. |
| chemical | Chemicals used in the experiment (e.g. buffer, salt, ligand). |
| reverse_transcriptase | The reverse transcriptase used to read out the chemical modification. |
| note | Extra notes about the data. |
| reactivity | Reactivity of the sequence. In a string format e.g. "[0.01,0.10,0.90,...]". |
| reactivity_err | Error of the reactivity. In a string format same as reactivity. |
RMDB S40 datasplit
In order to create the datasplit all sequences from RMDB_dataset_<integer>.parquet was collected along with RMDB_dataset_extra_clean.parquet. The first filter to was remove any sequences that only had chemical mapping data with signal to noise ratio < 1.0. Next, the sequences were all trimmed to remove the 5' and 3' ends of the sequence that had no chemical mapping data recorded. Finally, PDB sequences used in the tertiary structure analysis were also included for clustering. This resulted in RMDB_PDB.fasta.
Using mmseq2 the RMDB_PDB.fasta was clustered with 40% sequence identity. The results of the clustering is saved in RMDB_s40_clustering.tsv where the 1st column is the sequence ID cluster representative and the 2nd column is the sequence ID that got mapped to the cluster representative. The data was randomly split 20-80 for training and testing.
This results in the following files:
RMDB_s40_train.parquet: Train split containing the same data as the format above.RMDB_s40_train.fasta: Train split sequences.RMDB_s40_test.parquet: Test split containing the same data as the format above.RMDB_s40_test.parquet: Test split sequences.