EPCOTv2 Installation
====================

This document explains how to install and run **EPCOTv2**, the latest version hosted on Hugging Face Spaces, supporting inference on ATAC‑seq input to predict multiple genomic modalities in both human and mouse.

Dependencies
------------

- **Python 3.9+**
- `torch` (PyTorch) with GPU support recommended
- `huggingface_hub`, `gradio` (Space frontend)
- Standard Python libraries: `numpy`, `scipy`, `einops`, etc.

Installation Steps
------------------
Clone the Space repository
---------------------------

.. code-block:: bash

   git clone https://huggingface.co/spaces/luosanj/EPCOTv2
   cd EPCOTv2

Create and activate virtual environment
---------------------------------------

.. code-block:: bash

   python -m venv .venv
   source .venv/bin/activate

Install dependencies
--------------------

.. code-block:: bash

   pip install -r requirements.txt


Otherwise manually install core packages:

.. code-block:: bash

   pip install torch gradio huggingface_hub numpy scipy einops

Running Locally or in Colab
----------------------------

You can run the Gradio demo locally or on Google Colab:

- Local execution:

  .. code-block:: bash

     python app.py

- Colab:

  Open the Space and click “Open in Colab” (if enabled) to run interactive demo on GPU.

Usage Steps
-----------

1. Prepare an ATAC‑seq input file in `.pickle` format using provided scripts (e.g. `atac_process.py`).
2. Launch the demo either locally or via Colab.
3. Upload the `.pickle` file and specify genome (human/mouse) and modalities to predict.
4. Download outputs such as prediction archives and formatted `.bigWig` / `.bedpe` for visualization.

Notes
-----

- EPCOTv2 supports **both human (hg38)** and **mouse (mm10)** genomes.
- Internal scripts use **LoRA fine-tuned models** for efficient inference (see ``curriculum/loralib/utils.py``).
- The interface is provided through Hugging Face **Spaces** and is built using the **Gradio** framework.

References
----------

- Hugging Face Space: `luosanj/EPCOTv2 <https://huggingface.co/spaces/luosanj/EPCOTv2>`_
- Related methods described in the associated preprint and demo page: `bioRxiv preprint <https://www.biorxiv.org/content/10.1101/2025.05.08.652986>`_