rouskinhf 0.4.8

A library to manipulate data for our DMS prediction models.

Download your RNA data from HuggingFace with rouskinhf!

A wrapper around Huggingface the load data for eFold. You can:

• pull datasets from the Rouskinlab's HuggingFace
• create datasets from local files

Installation

To download data

pip install rouskinhf

To push data to huggingface (optional)

• get a token access from the rouskilab huggingface's page
• add this token to your environment

export HUGGINGFACE_TOKEN="hf_yourtokenhere"

To predict structures from rouskinhf (optional)

You'll need to install D. Mathew's RNAstructure Fold (also available on Rouskinlab GitHub).

Check your RNAstructure Fold installation in a terminal:

Fold --version

How to use

Download a dataset

import rouskinhf

rouskinhf.get_dataset(
    name='bpRNA-1m', # the name of a dataset from huggingface/rouskinlab
    force_download = False # use a local copy of the data if it exists
)

Convert whatever format to rouskinhf format

import rouskinhf

rouskinhf.convert(
    format = 'ct', # can be ct, seismic, bpseq, fasta or json (rouskinhf output data structure)
    file_or_folder = 'path/to/my/ct/folder',
    predict_structure = False, # Add structure from RNAstructure
    filter = True, # removes duplicates, non-regular characters and low AUROC
    min_AUROC=0.8,
)

Note: Sequences with bases different than A, C, G, T, U, N, a, c, g, t, u, n are not supported. The data will be filtered out.

Rouskinhf structure format

# rouskinhf_output_file.json
{
    "reference_name": {
        "sequence": "CACGCUAUG",
        "structure": [(0,8), (1,7)], # base pair representation
        # whatever other info you need
    }
}