janggu_usecases

Examples for deep learning in genomics using Janggu

Requirements

jupyter
bedtools
pybedtools
samtools
dash
janggu
R
rpy2
tzlocal
r-ggplot2
r-ggrepel
r-dplyr
statsmodels
pandas
numpy

These can be installed via conda and pip.

The respective cells in the notebook for installing requirements may be outcommented.

Download the datasets

In order to download the required datasets, enter the 00_preparation folder. It contains jupyter notebooks that specify and control the data download. Furthermore, it sets up the regions of interest for the model training and evaluation.

Note

Some of the steps in the notebooks may be outcommented or deactivated, including the invocation of time-consuming training steps, so that during evaluation, they are not re-run. You may either activate them within the notebook or invoke the scripts on the command line if you wish to train the models from scratch. It may also be necessary to adapt the use of CUDA_VISIBLE_DEVICES (see tensorflow docs). The GPU device is selected via the -dev option in use case 2. These were chosen for our specific setup with 8 GPUs. For example, if you only have access to one GPU specify CUDA_VISIBLE_DEVICES=0 before running the scripts.

JunD prediction

Run the jupyter notebook 'predicting_jund_binding.ipynb' in order to reproduce the results. You can control on which gpu the models are trained by specifying the environment variable CUDA_VISIBLE_DEVICES (see tensorflow documentation).

DeepSEA and DanQ experiments

To train and evaluate the DeepSEA and DanQ comparison, enter the '02_deepsea_danq_prediction' folder and launch the jupyter notebook 'deepsea_danq_experiments.ipynb'. To activate model training, set the parameter train_models = True. Otherwise, the notebook merely evaluates the results. You may need to adapt -dev to select a specfic GPU.

CAGE-tag prediction

To reproduce the CAGE-tag prediction use case, enter '03_cage_prediction' and launch the 'predicting_cage_tags.ipynb' notebook. In order to run the cross-validation analysis, outcomment the respective command line invocations of the script 'cage_prediction.py'. You can control on which gpu the models are trained by specifying the environment variable CUDA_VISIBLE_DEVICES (see tensorflow documentation).