Patches

Patches#

Tests Code style: black Docs License: MIT Code style: black

https://github.com/Computational-Morphogenomics-Group/Ladder/blob/main/repo_resources/logo.png?raw=true

Patches takes as input a collection of observations or gene expression profiles obtained from multiple conditions referred to as groups – for example, age, treatment or injury stage. Each group is characterized by a unique set of categories or attributes – for example, ‘young’ or ‘old’ represent attributes of the group ‘age’. Given an anndata object with metadata for the observations contained in separate columns of anndata.obs, Patches can:

  • Learn conditional and condition agnostic latent representations for observations.

  • Transfer observations across combinations of conditions.

  • Provide attribute specific scores for each gene, associating gene expression with effects of attributes.



Table of Contents#

  1. Installation

  2. Repository structure

    1. Workflow API (for general users)

    2. Developer API (for advanced users)

  3. Getting started

  4. Release notes

  5. Issues and contact

  6. License and citation

Installation#

  • You will need Python 3.10 or newer in your system. If you don’t have a Python environment setup, we recommend installing it through Mambaforge.

  • Install everything with pip, either through PyPI or GitHub:

pip install scladder

pip install git+https://github.com/Computational-Morphogenomics-Group/Ladder.git

Repository Structure#

Alongside Patches, this repository provides the low and high-level API’s (called Ladder, still in development at time of Patches release) for running similar models with data from multiple conditions built on anndata.

  • All code is contained under the src directory. (add modules and stuff here)

  • Tests are contained under the tests directory.

  • All documentation can be generated and is serviced through the docs directory.

Workflow API#

The high-level workflow API provides a very easy interface to apply models considered in the study (scVI, scANVI, Patches; all built on the Pyro framework) to multicondition datasets easily.

  • This is recommended for users that want to directly apply the model, without worrying much for technical details.

  • Consult the workflow docs for hyperparameters you can define and how to define them.

from ladder.data import get_data
from ladder.scripts import CrossConditionWorkflow, InterpretableWorkflow # Workflow APIs, read docs on which one you'd like for your case!

# Initialize workflow object
workflow = InterpretableWorkflow(anndata, verbose=True, random_seed=42)

# Define the condition classes & batch key to prepare the data
factors = ["time", "age", "broad_type"]
workflow.prep_model(factors, batch_key="sample", model_type='Patches', model_args={'ld_normalize' : True}) # Define model and optimizer hyperparams

# Train the model
workflow.run_model(max_epochs=2000, convergence_threshold=1e-5, convergence_window=2000) # Define convergence hyperparams

Developer API#

The low-level developer API explicitly performs all actions condensed into workflows. This allows for:

  • Implementing and running additional models using the Pyro framework.

  • Data preparation functions to provide ease for running the models and downstream visualizations.

  • Evaluation of generative accuracy alongside mixing/separation of conditional effects using defined metrics.

This API is designed to be specifically used by those who are interested in specialized applications or those who would like to extend the codebase. Please use the issue tracker for any questions / encountered problems.

Getting started#

Please check out the documentation for all instructions and tutorials.

Release notes#

See the changelog.

Issues and contact#

For additional questions, please reach out to:

  • ob2391@columbia.edu

  • bmd2151@columbia.edu

If you encounter any bugs, please report it through the issue tracker. In general, providing a minimally reproducible example is the best way to facilitate easy debugging of issues.

License and citation#

This project is licensed under the terms of the MIT License. If the codebase has been of any use to you, please cite:

Beker, O., Amador, D., Nima, J., Deursen, S., Woappi, Y., & Dumitrascu, B. (2024). Patches: A Representation Learning framework for Decoding Shared and Condition-Specific Transcriptional Programs in Wound Healing. bioRxiv.

BibTeX:

@article {Beker2024.12.23.630186,
	author = {Beker, Ozgur and Amador, Dreyton and Nima, Jose Pomarino and Deursen, Simon Van and Woappi, Yvon and Dumitrascu, Bianca},
	title = {Patches: A Representation Learning framework for Decoding Shared and Condition-Specific Transcriptional Programs in Wound Healing},
	elocation-id = {2024.12.23.630186},
	year = {2024},
	doi = {10.1101/2024.12.23.630186},
	publisher = {Cold Spring Harbor Laboratory},
	URL = {https://www.biorxiv.org/content/early/2024/12/24/2024.12.23.630186},
	eprint = {https://www.biorxiv.org/content/early/2024/12/24/2024.12.23.630186.full.pdf},
	journal = {bioRxiv}
}