nrc: A Python repository from IcarPA-TBlab

nRC: non-coding RNA Classifier

A tool for classification of ncRNA sequences based on structural features extracted from RNA secondary structure and a deep learning architecture implementing a convolutional neural network.

Dependences:

Python (2.7.5)
Theano (0.8.2)
NumPy (1.11.2)
Pyyaml (3.12)
Java JDK (1.8.0)
ViennaRNA (1.8.5)
Ipknot (0.0.2)
MoSS (2.13)
GLPK (4.60)

nRC tool docker image:

https://hub.docker.com/r/tblab/nrc/

How to Use

The best way to use nRC tool is to create a container from the docker image above. Anyway, it is possible to download source code, install all dependences and execute the following steps.

nRC Training process

Step 1: training feature model

nrc_training_feature_model.sh -d <nRNA_training_file>.fasta -o <experiment_name> -n <graph_feature_max_size> -m <graph_feature_min_size>

Step 2: traning CNN (create classification model)

nrc_training_network_model.sh -d <experiment_name>_<graph_feature_max_size>_<graph_feature_min_size>.txt -p <parameters>

nRC Testing process

Step 3: testing feature model

nrc_testing_feature_model.sh -d <nRNA_testing_file>.fasta  -f <experiment_name>_<graph_feature_max_size>_<graph_feature_min_size>.nel -o <sequence_output_name>

Step 4: testing CNN (test classification model)

nrc_testing_network_model.sh -d <sequence_output_name> -p <parameters> -m <experiment_name>_<graph_feature_max_size>_<graph_feature_min_size>.pkl -o <classification_output_name>

Fasta file format

Note that each sequence in fasta format should have the following header:

>seq_ID class_name

wher seq_ID and class_name labels must be without ".","","/" or any special character, e.g.:

>RF00005_AAFR03000905_1_148681-148750 tRNA
AGCAGTGTGGCATAGTGGAAAGTGTTGGATTTGTAGTTAAAGGACTTGGGTTCAGATCCC
TGCTCTGTTA

Datasets

This distribution contains two datasets. Both of them are available in "data" folder.

The "sample" dataset is a small dataset with 40 ncRNA fasta sequences (belonging to 4 ncRNA classes) for training and 20 ncRNA fasta sequences for testing.
The "paper" dataset has been used for experiments in the manuscript submitted for pubblication at international journal. It is composed by a training dataset with 6320 ncRNA fasta sequences (belonging to 13 ncRNA classes) and two validation datasets with respectively 2600 fasta sequences (13 classes) and 2400 ncRNA fasta sequences (12 classes). All of them are extracted from Rfam database release 12. nRC trained models used in the manuscript submitted for pubblication at international journal are available at http://tblab.pa.icar.cnr.it/public/nRC/paper_dataset/

IcarPA-TBlab/nrc