FISHFactor: A Probabilistic Factor Model for Spatial Transcriptomics Data with Subcellular Resolution
Code repository supplementing the paper.
FISHFactor is a non-negative, spatially informed factor analysis model with a Poisson point process likelihood to model single-molecule resolved data, as obtained for example from multiplexed fluorescence in-situ hybridization methods. In addition, FISHFactor allows to integrate multiple cells by jointly inferring cell-specific factors and a weight matrix that is shared between cells. The model is implemented using the deep probabilistic programming language Pyro and the Gaussian process package GPyTorch.
- src/ contains the FISHFactor model, data simulation and util functions.
- experiments/ contains the experiments described in the paper.
- data/ contains scripts to download and process data used in the paper.
The required packages can be installed in an Anaconda environment using the environment.yml file. An example for using FISHFactor with simulated cells is shown in fishfactor_demo.ipynb.