Fast and accurate microRNA search using CNN

Requirements

• Python: 3.6
• Keras: 2.2.4
• viennarna: 2.4.11
• Other python modules: numpy, matplotlib, scikit-learn, pandas, biopython, opencv, tqdm

Usage

You can see the valid parameters of the script by help option:

$ python <SCRIPT_FILE> --help

• Clone the repository or download compressed source code file.

$ git clone https://github.com/HubertTang/DeepMir
$ cd DeepMir

• Generate input file for CNN:

  INPUT: The directory containing the original .fasta files of different families should be saved in data/:

  data/
  |---rfam_mirna/
    	|---RF00027.fa
  	|---RF00047.fa
    	|---RF00051.fa
  		...

  OUTPUT: The directory containing generated files will be save in inputdir/ by default. The default name of the directory is like: 165_60_onehot . 165 is the number of the classes in this dataset, 60 is the number of sequences of each family, onehot is the encoding method.

  • For pair/ probability/ mixed matrix method:

  One example of parameter setting for generating Rfam-60 using pair matrix:

  $ python fasta2img.py 60 pair \
    --fasta_dir rfam_mirna \
    --max_length 200 \
    --split_ratio 6 \
    --threshold 0 \
    --coding 1 \
    --output_dir inputdir \
    --mode 2 \
    --shuffle True

  • For one-hot matrix method:

  One example of parameter setting for generating Rfam-60 using one-hot matrix:

  $ python fasta2onehot.py 60 \
    --fasta_dir rfam_mirna \
    --max_length 200 \
    --split_ratio 6 \
    --output_dir inputdir \
    --mode 2 \
    --shuffle True

• Train CNN model and evaluate the performance:

  INPUT: The input file saved in inputdir/.

  OUTPUT: After training, the performance of the model will be evaluated automatically using the testing data. The trained model and the performance will be saved in a directory, named by data, in log/.

  Use train_img.py to train the pair/ probability/ mixed matrix model, use train_onehot.py to train the one-hot matrix model.

  One example of parameter setting for training pair matrix model using dataset from 165_60_pair

  $ python train_img.py 165_60_pair 165 \
    --seq_length 200 \
    --learning_rate 0.001 \
    --batch_size 32 \
    --num_epochs 20 \
    --filter_sizes 2 2 \
    --num_filters 32 64 \
    --keep_prob 0.5 \
    --num_hidden 128 64 \
    --loss_function categorical_crossentropy \
    --optimizer Adam \
    --log_dir log

Data

All datasets can be download from HERE. Please read the README.txt before using.

Contact

If you have any question or problem, please feel free to send email to xubotang2-c@ad.cityu.edu.hk