RNA State Inference with LSTMs

Devin Willmott, Dr. David Murrugarra, Dr. Qiang Ye

---

License

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program. If not, see http://www.gnu.org/licenses/.

Required Software and Packages

• python 3.6
• numpy 1.14.3
• keras 2.1.5
• theano 0.9.0 or tensorflow 1.4.1
• gtfold 1.16
• (Optional) CUDA 8.0 and cuDNN 5.1

Instructions

1. Clone this repository, install all required packages above.

2. Get data, available at [http://www.ms.uky.edu/~dmu228/rna_lstm/rnastateinf-data/]. This will contain training and testing data for the neural network and HMM in the 'data' directory, and the output of the neural network used to direct NNTM in the 'probabilities' directory. Place these into the cloned directory.

State Inference

Type

python rnn.py

to begin training and testing a neural network with the architecture and hyperparameters from the paper, or use the command line arguments to choose your own. For example, to train and test an RNN with a learning rate of 0.01 and a network with three bidirectional LSTM hidden layers of sizes 300, 200, and 100, run

python rnn.py --lr 0.01 --hiddensizes 300 200 100

The HMM has two command line arguments: the mode, and k, the order of the HMM. There are three modes: 'train' trains and saves an HMM of order k, 'run' uses a saved HMM of order k to perform inference on the test set, and 'cycle' does train' and 'run' using all orders of HMM from 1 to k. So, to train an order 4 HMM, type

python hmm.py 4 train

Predicted State Directed NNTM

The method.py file takes the output of the neural network and uses it to perform state directed NNTM. With the neural network output in the directory names 'probabilities', running

python method.py

will generate SHAPE and produce three predicted structures for each sequence: no SHAPE direction, native state direction, and predicted state direction. It will print PPV, sensitivity, and accuracy of all three predictions.