This tool is developed for circRNA-binding event identification from CLIP-seq data using deep temporal convolutional network
HCRNet is written in Python3 and requires the following dependencies to be installed:
We recommend you to build a python virtual environment with Anaconda.
conda create -n hcrnet python=3.7.6
conda activate hcrnet
https://github.com/yangyn533/HCRNet.git
cd HCRNet
The pre-trained models (including circRNA2Vec_model, linRNA2Vec_model and bert_model), the 37 circRNA fragment datasets, 31 linear RNA datasets, a full-length circRNA dataset containing 740 sequences and the eCLIP data with binding sites of 150 novel RBPs can be downloaded in this Repositories. Meanwhile, all supporting datasets and source codes for our analyses are also freely available at (https://doi.org/10.6084/m9.figshare.16943722.v1).
python HCRNet-Train.py [--RBP_ID <circRNA or linearRNA or eCLIP data>]
[--kmer <default=3>]
[--modelType <default='~./Pre-trained models/circRNA2Vec_model'>]
[--storage <default='~./resultpath/result/'>]
You can train the model of 5-fold cross-validation with a very simple way by the command blow: Python HCRNet-Train.py.
The script of if name == "main calls training process which trains several models of each model type for an RNA and finds the best set of hyperparameters. The main function then trains the models several times (num_final_runs) and saves the best model.
The HCRNet-Predict.py is proposed to calculate the probability for the circRNAs of unknown types. Please also change following paths to suit your system:
-
set the sequence location. e.g.,
seqPath = '/home/Sequence/'
-
set the type of the RNA Embeddings. e.g.,
modelType = '/modelpath/circRNA2Vec_model'
-
set the type of circRNA model. e.g.,
modelPredictType = '/finalmodel_path/model.h5'
The prediction results will be displayed automatically. If you need to save the results, please specify the path yourself.
We also provide a website http://39.104.118.143:5001/. HCRNet provides identification of the specific binding events for circRNA and linearRNA segments or full-length circRNA sequences. Meanwhile, HCRNet also allows users to facilitate the identification of potential circRNA-RBP binding targets with a generic strategy model.
Thank you and enjoy the tool! If you have any suggestions or questions, please email me at yangyn533@nenu.edu.cn.