MASS

MASS is a deep learning framework for multi-species RNA modification analyses by integrating RNA modification data from multiple species. This is an instruction of predicting RNA modification using MASS.

Dependencies

This repository has been tested on Ubuntu 16.04. We strongly recommend you to have Anaconda3 installed, which contains most of required packages for running this model.

Must installed packages or softwares

tensorflow 1.8.0
tensorlayer 1.10.1
pandas 0.12.0
CD-HIT

Data Preparation

Download complete data from the RMBase 2.0 or ENSEMBL, and unzip these data to ./data/YOUR_DATA_NAME.
You can preprocess your own data with ./src/pocessing_pipeline.py, and move the processed data to ./data/YOUR_DATA_NAME.
For the training process, you should store positive samples in data/YOU_DATA_NAME/positive_samples/train and negative samples in ./data/YOUR_DATA_NAME/negative_samples/train.

Get Started

Test with pre-trained models

The trained multi-speices model were stored in ./checkpoint
You can run the script ./src/main.py to generate prediction on test data, and the results will be stored in ./predict_result. It will also report AUPRC and AUPR based on the given data and label. Sample code:

cd src
python main.py --name mass --data full_data --sn 8 --mode test --fold 0 --gpu 0

Train new models

You can also train mass with you own data:

cd src
python main.py --name mass --data full_data --sn 8 --mode train --gpu 0

Utility files

data2fa.py is used to convert dataset into fasta format

main.py is the main function that has the following arguments:

--name: experiment name
--gpu: gpu id, currently MASS can only run on single GPU
--data: data source for trainning for
--sn: number of species
--fold: index for species in ./src/config.sample_names

mlcb-thu/MASS