README¶

Version: “v0.1.0”

Author: “Zijun Zhang”

Date: “2.17.2019”

Table of Contents¶

Installation
Testing
Using Darts DNN
Training from scratch

Installation¶

Install via Anaconda¶

The recommended way to install Darts_DNN is through Anaconda. You can also create a new environment for Darts, because currently DARTS works in Python 2.7.

conda create -n darts python=2.7 # optional
source activate darts
conda install -c darts-comp-bio darts_dnn

This will allow conda to do all the heavy-lifting and most often the easiest way to get things spinning.

From GitHub¶

Alternatively, to install Darts_DNN python package from Github, navigate to this folder, then type

> cd Darts_DNN
> make install

There are a few Deep-learning packages that Darts_DNN requires, including the popular high-level interface Keras.

Testing¶

To test whether you have successfully installed Darts_DNN, type the following command in your shell:

> Darts_DNN -h
usage: Darts_DNN [-h] [--version] {train,predict,build_feature,get_data} ...

Darts_DNN -- DARTS - Deep-learning Augmented RNA-seq analysis of Transcript
Splicing

positional arguments:
  {train,predict,build_feature,get_data}
    train               Darts_DNN train: train a DNN model using Darts
                        Framework from scratch
    predict             Darts_DNN predict: make predictions on a built feature
                        sets in h5 format
    build_feature       Darts_DNN build_feature: build feature file given
                        required information
    get_data            Darts_DNN get_data: connects online to get Darts_DNN
                        data for the current version.

optional arguments:
  -h, --help            show this help message and exit
  --version             show program's version number and exit

For command line options of each sub-command, type: Darts_DNN COMMAND -h

Most often one will be using the predict subcommand to make predictions. Using the test_data download from Github, you can test whether this function works properly:

cd test_data/
tar -xvzf A5SS.thymus_adipose.tgz
Darts_DNN predict -i darts_bht.flat.txt -e RBP_tpm.txt -o pred.txt -t A5SS

The “A5SS.thymus_adipose.tgz” is the Roadmap thymus-adipose tissue-specific Alternative 5’ splice sites analysis results generated by Darts BHT. Note that all thymus-related comparisons, including this one, is held-out data and never seen by the trained Darts_DNN model.

Using Darts DNN¶

For more details, please refer to the documentation site at ReadTheDocs here. Below we provide a minimal example.

Assume you have already run Darts_BHT and get the Darts-flat inference output file, say "darts_bht.flat.txt". There are two simple steps to run Darts_DNN prediction on it:

Darts_DNN build_feature¶

You will need to build the feature file for your target Darts-flat output. The input is "darts_flat.out.txt", and the output is a feature set in hdf5 data store. Below is an example:

Darts_DNN build_feature -i darts_flat.out.txt \
-c /path/to/ENCODE_sequenceFeature_absmax_normalized.h5 \
-e /path/to/condition1/kallisto/ /path/to/condition2/kallisto/ \
-o data.h5

Darts_DNN predict¶

Now run the predict. This will also estimate the prediction accuracy by the significant Darts-flat events, so that the users can decide whether to proceed running Darts_BHT with the informative prior.

Darts_DNN predict -i data.h5 -o pred.txt \
-m /path/to/model_param.h5