Compatibility with scanpy ecosystem in python

[linjing-lab]

Hi, I have developed some tools in spatial transcriptomics and spatial multi omics, such as stkan and mmspao, both of which are based on the scanpy ecosystem. Therefore, I find it relatively easy to read many spatial clustering tool architectures. But when I saw the SpatialExperiment project in the BiocPy community today, I was a bit unfamiliar with using Bioconductor's data format. I want to know what tools are needed for a complete downstream analysis task of spatial transcriptomics in BiocPy, including SpatialExperiment.

Then I want to know if BiocPy can completely replace Bioconductor now. I am familiar with graph algorithms for spatial omics such as scanpy and pytorch, but I want to familiarize myself with the process of Bioconductor and learn if BiocPy can be sufficient for delivery.

[jkanche]

for the first part of the question, we provide methods to convert SpatialExperiment to anndata from the scverse ecosystem (reference) and vice-versa. The coercion function was originally written for the python implementation of voyager methods, details discussed in the reference link. These are however lossy since anndata doesn't have all the slots that bioconductor representations usually contain.

BiocPy is not a replacement of Bioconductor. It is an effort to bring some of the well-established data structures and methods from R to Python, for ease of use with the broader python ecosystem. check out how libscran and singler use the SingleCellExperiment data structure and support single-cell analysis. Let us know if you have any questions or feedback. Happy to help if you run into issues.

[LTLA]

Also to assist interoperability between R and Python, also see rds2py and dolomite-base for some examples of (mostly) lossless exchange formats that are enabled by BiocPy data structures.

[linjing-lab]

I am currently conducting research on techniques and methods related to spatial transcriptomics, such as spatial perturbation methods such as River, Celcomen, and CONCERT, as well as synergistic information cell_synergy for foundation models. They are all high-level PyTorch code for Scanpy and Anndata, and my Linux environment has a complete set of dependency libraries for Scanpy and PyTorch. I can contribute to single-cell and transcriptomics tools of the Python ecosystem, and if I can join BiocPy to developing downstream task, my ability on bioinformatics software development can improve.

[jkanche]

We don't have any official process of joining BiocPy. It is an open source effort so feel free to develop your packages that use these data structures or methods. If it helps, checkout the developer guide for some established practices around python package development, mainly using biocsetup. If something is missing or you run into bugs, let us know.

[jkanche]

To add, it might be worth joining the bioconductor zulip channel and the spatial groups for finding collaborators and folks working on these modalities.

[linjing-lab]

Thanks. I am now more concerned about the foundation model. I am thinking of using existing tools to build a foundation model from scratch that can perform embed, integrate, annotate, and perturbation tasks.