asdeep: Deep-learning tool to interpret variatn function by allelic imbalance.

Introduction

Dependencies

The tool was written in Python 3 and depends on a set of well-developed packages, mainly including the followings:

• numpy == 1.21.4
• h5py == 3.6.0
• pymc3 == 3.11.4
• pysam == 0.18.0
• torch == 1.10.0
• captum == 0.4.1
• tensorboard == 2.7.0
• torchvision == 0.11.1
• scikit-learn == 1.0.1

Installation

We highly recommend to use a Python virtual environment. Here is an example to create one by the venv module shiped with Python 3.

$> python -m venv .env
$> source .env/bin/activate

Then, clone the repository and install it using pip module.

(.env) $> git clone https://github.com/zhenhua-zhang/asdeep
(.env) $> python -m pip install -e .

Or you can install it from PyPi.

(.env) $> python -m pip install asdeep

Usage

Example dataset

We also provide a small dataset for (potential) users to play with. The dataset is permantly available by the link at Zenodo

Step-by-step usage example

Create a working space

Prepare input files

Prepare the allelic read counts

Estimate the ASE effects for an isoform

Create databases for train/test/prediction

Train a model and test it

Predict the unseen samples

Outputs

• Model state
• Evaluation matrix
• Prediction results
• Feature attribution plot

Known issues

1. No section: 'blas' problem raised by pymc3

The pymc3 package depends on Theano-PyMC which is a special Theano branch. However, one knownd issues when importing pymc3 is the No section: 'blas'. if this is the issues you come across, please check if the BLAS installed correctly in your system.

Contact

If you have issue or question, please create a new issue or send an email to the following address.

Zhenhua Zhang zhenhua.zhang217@gmail.com