AgentBind is a deep-learning framework that analyzes gene-regulatory regions in the human genome and identifies genomic positions with strong effects on regulatory activities. This Github repository contains code for an experiment in which we apply AgentBind to analyzing H3K27ac signals in brain cells.
System Requirement & Installation | Data Download | Data Preparation | Run Experiments | Demo | Contact
Code in this repository has been tested on CentOS Linux 7 (core) with Python (v3.7.4) in the Anaconda (v4.11.0) environment. Please make sure you have installed the following python libraries before you run the code.
Our code requires external python libraries including Tensorflow v2.3.0 , biopython v1.71, numpy v1.18.5, six v1.15.0, and scikit-image v0.21.3. You can install them with the conda package manager.
Input data of the experiments can be downloaded using this link: download
data_prep.py is a python script that prepares training/validation/test data for the deep learning models and converts DNA sequences into the one-hot encoding format.
Required parameters:
- --path: path to the data folder that you download from our Google Drive link.
To run this script, you can simply execute:
python data_prep.py --path {your-data-path}/storage
run_experiment.py is a python script that trains a deep learning model for each brain cell type and annotates all the bases in the input data with importance scores.
Required parameters:
- --path: path to the data folder that you download from our Google Drive link.
To run this script, you can simply execute:
python run_experiment.py --path {your-data-path}/storage
Once the program finishes, you can find the classification results in folder {your-data-path}/storage/experiments/results_extended_coordconv_sliding/{cell-type}/category_0/. And the annotation results can be found in folder{your-data-path}/storage/experiments/results_extended_coordconv_sliding/{cell-type}/annotations/
This python program "run_experiment.py" takes ~24-48 hours to complete. If you need the Grad-CAM annotation scores only, you can directly download them here: download
To acquire importance scores for H3K27ac regions in brain cells, you can either follow the steps above or download them directly using the links below:
- Astrocyte: https://drive.google.com/file/d/1RB5Nx6k4kxcnNWilBnWISeGcFZa6nkxQ/view?usp=sharing
- Neuron: https://drive.google.com/file/d/1_YxH7KfLmbiSpJyVo0FJUxAyly2zsbiG/view?usp=sharing
- Microglia: https://drive.google.com/file/d/1a0RskxIOvNRSiUTBFUyD8XT78Qy9i18J/view?usp=sharing
- Oligdendrocyte: https://drive.google.com/file/d/1BkxXk-hMHRuV9652o0G5Xyy-nEa1Qle2/view?usp=sharing
You can view these tracks through IGV: https://tinyurl.com/ya2rc6nu
For questions on usage, please open an issue, submit a pull request, or contact Melissa Gymrek (mgymrek AT ucsd.edu) or An Zheng (anz023 AT eng.ucsd.edu).