mNSF

This GitHub repository is associated with our paper available on bioRxiv: https://www.biorxiv.org/content/10.1101/2024.07.01.599554v1

All code to analyze the data and generate figures is available at https://github.com/hansenlab/mNSF_paper

Tutorial for using mNSF on example datasets are publicly available at https://github.com/hansenlab/mNSF/blob/main/tutorial/dlpfc.md (use DLPFC data as example) and https://github.com/hansenlab/mNSF/blob/main/tutorial/mouse.md (use mouse saggital section data as example).

Demo notebooks of mNSF using simulated data

1. Getting Started with mNSF

   • Basic end-to-end workflow using synthetic data
   • Data preparation, model training, and result visualization
   • Parameter selection guidance

2. Large Dataset Optimization

   • Techniques for handling large-scale spatial transcriptomics data
   • Optimizing induced points selection
   • Memory and computational efficiency strategies

3. Effect of Parameter Selection

   • In-depth analysis of how different parameters affect mNSF performance
   • Interactive visualizations of parameter interactions
   • Practical guidelines for parameter tuning

4. Moran's I Tutorial

   • Understanding spatial autocorrelation in mNSF factors
   • Calculating and interpreting Moran's I statistics
   • Visualizing different levels of spatial structure

5. Number of Factors Selection

   • Statistical approaches for determining optimal factor count
   • Using deviance explained and elbow methods
   • Practical guidelines for different dataset complexities

6. Factor Interpretability Guide

   • Connecting mNSF factors to biological meaning
   • Techniques for visualizing gene-factor relationships
   • Step-by-step interpretation framework

Installation

You can install everything from the PyPI repository using pip install -e . but Tensorflow will most likely not install. A safer way would be to use conda to setup most of the packages then use pip to install.

Install using pip

1. Git clone and activate your environment of choice.
2. Install tensorflow.
3. pip install -e .

Install using conda/mamba

1. Git clone this repo git clone https://github.com/hansenlab/mNSF/ and enter cd mNSF.
2. Install conda. I recommend this distribution: https://github.com/conda-forge/miniforge. Do not install the full anaconda, it's highly bloated.
3. Create a new environment and install using

conda env create -n mnsf -f environment.yml
conda activate mnsf

The package should be available right away.

3. Install tensorflow.

CPU only

conda install tensorflow

GPU

If you have a GPU and is operating in a Linux system, you can in the `mnsf` environment.

conda install tensorflow-gpu

Development

This package is managed by twine. Assuming twine is installed in your python version, you build the distribution by

python setup.py sdist

inside the repository directory, and then you upload to PyPI by

twine upload dist/*

(requires an account on PyPI)