mRNN - mRNA RNN (mRNA recurrent neural network)

mRNN is a package for distinguishing coding transcripts from noncoding using gated recurrent neural networks (GRNNs).

How to get mRNN

You can obtain mRNN through github:

git clone https://github.com/hendrixlab/mRNN.git

Note: This code requires the module "Passage" available at: https://github.com/IndicoDataSolutions/Passage However, it is tested to work properly on an early version that can be obtained by:

git clone https://github.com/IndicoDataSolutions/Passage.git
cd Passage
git checkout 4b8be6dc4d17ccc78a21abc82afb2d0f72c04b91

Also required are:

numpy
theano
cPickle
sys
os
math

mRNN: Included Files

1. Testing and Training

   train_mRNN.py - training mRNN model based on input training sequences.

   Example: $ python mRNN/train_mRNN.py -e 128 -r 32 -d 0.4 -o mRNN.16K.da1.1 -s 3 -E 4 mRNAs.train16K.da1.fa lncRNAs.train16K.da1.fa mRNAs.valid500.fa lncRNAs.valid500.fa > epochs.mRNN.16K.da1.1.txt

   test_mRNN.py - test the accuracy of a known set of mRNAs and lncRNAs.

   Example: $ python mRNN/test_mRNN.py -w w14u3.pkl -f test mRNAs.test500.fasta lncRNAs.test500.fasta

   test_ensemble_mRNN.py - Ensemble testing with mRNN models. Models are provided as a comma-separated list.

   Example: $ python mRNN/test_ensemble_mRNN.py -w w10u5.pkl,w14u3.pkl,w16u5.pkl,w18u1b.pkl,w23u2.pkl -f test mRNAs.test500.fasta lncRNAs.test500.fasta

   mRNN.py - Basic prediction of coding probability for a set of input sequences.

   mRNN_ensemble.py - Basic prediction of coding probability for a set of input sequences using a set of models.

2. Further Analysis

   mutation_analysis.py - For the provided input sequence, perform a mutation analysis involving every possible point-mutation on the input sequence, and computing its corresponding score change.

   pair_mutation_analysis.py - For a provided input sequence, computer all possible pairs of mutation i,a,j,b where position i is changed to a and position j is changed to b.

   shuffle_analysis.py - For the provided input sequences, shuffle each of them and report the scores of the shuffled sequence to the unaltered sequence.

   truncation_analysis.py - Coding score trajectory. Compute the mRNN coding probability for all truncations of the input sequence from position 1 to i.

3. Core mRNN modules:

These modules aren't called directly, but involved in basic function of mRNN

  fasta.py - input reading FASTA files

  model.py - building the mRNN model

  evaluate.py - Computing accuracy, batch testing

  preprocessing.py - Utilities for various preprocessing

How to get data from mRNN project

The data used to train mRNN and produce the weights provided in the weights directory, you can download them from the following location:

https://osf.io/4htpy/

262K lncRNAs.CHALLENGE.fa

243K lncRNAs.CHALLENGE500.fa

23M lncRNAs.MOUSETEST.fa

977K lncRNAs.TEST.fa

527K lncRNAs.TEST500.fa

28M lncRNAs.TRAIN.fa

11M lncRNAs.train16K.fa

981K mRNAs.CHALLENGE.fa

708K mRNAs.CHALLENGE500.fa

146M mRNAs.MOUSETEST.fa

5.0M mRNAs.TEST.fa

1.2M mRNAs.TEST500.fa

191M mRNAs.TRAIN.fa

12M mRNAs.train16K.fa