A deep neural network for predicting SHAPE reactivity scores from high-throughput RNA probing data.

Requirements:

• Python version 3.6.8
• TensorFlow version 1.13.1
• keras version 2.2.4
• numpy version 1.16.2

Command line for training and testing:
python main.py <train/test> <sequences/RNAplfold/both> <binary_crossentropy/mse> <in_vitro/in_vivo> <training/testing sequences file> <training/testing RNAplfold file> <training/testing annotation file>

Inputs:

• <train\test> - Train or test the network

  • "train" - Train DeepSHAPE network.
  • "test" - Predict icSHAPE rectivity scores using DeepSHAPE trained netork.

• <sequences\RNAplfold\both> - Controls the input data type

  • "sequences" - Select only sequence data as an input to the network
  • "RNAplfold" - Select only RNAplfold data as an input to the network
  • "both" - Select both sequence data and RNAplfold data as an input to the network

• <binary_crossentropy\mse> - Controls loss function selection and and its corresponding activation function of the last layer

  • "binary_crossentropy" - Binary cross entropy loss function and sigmoid activation function of the last layer.
  • "mse" - Mean squared error loss function and linear activation function of the last layer.

• <in_vitro\in_vivo> - Controls dataset selection

  • "in_vitro" - In vitro dataset
  • "in_vitro" - In vivo dataset

• <training/testing sequences file> - Path to training sequences file

• <training/testing RNAplfold file> - Path to training RNAplfold file

• <training/testing annotation file> - Path to training SHAPE file
  

Note - It is required to provide three valid input files regardless of the chosen input data type.

Outputs:

• Training Outputs stored in outputs/saved_models/
  • log.txt
  • .hdf5 - Trained network for each epoch
  • acc_history.txt - Accuracy value after each epoch
  • loss_history.txt - Loss value after each epoch
• Testing Outputs stored in outputs/test_results/
  • log.txt
  • predictions_.txt - SHAPE predictions (one file per epoch)
  • performance.txt - Network performance

Note - The outputs will be stored in a path that indicates the chosen training configuration under 'outputs' folder, i.e - (1)Dataset selection (2) input data type selection (3) loss function selection