Interoperability between Bioconductor and Python for scRNA-seq analysis

Unlike traditional bulk RNA-seq analysis which is dominated by
Bioconductor, packages for analysing single-cell RNA sequencing data are more fragmented. Currently, there are three key ecosystems, the
Seurat package (https://satijalab.org/seurat/) (available from CRAN), Bioconductor's SingleCellExperiment object (https://bioconductor.org/packages/SingleCellExperiment/) and the AnnData Python object (https://anndata.readthedocs.io/en/latest/index.html) used by the Scanpy package (https://scanpy.readthedocs.io/en/stable/). While these platforms each have strengths and weaknesses most analysts are likely to only use one of them. In this talk, I discuss how interoperability between R and Python can allow us to take advantage of these platforms strengths
and avoid unnecessary reimplementation of methods. I highlight the reticulate R package for interacting with Python (https://rstudio.github.io/reticulate/), the basilisk package for encapsulating Python environments (https://bioconductor.org/packages/basilisk/), my zellkonverter package for
converting between AnnData and SingleCellExperiment objects (https://bioconductor.org/packages/zellkonverter/) and the velociraptor package as an example of wrapping a Python package (https://bioconductor.org/packages/velociraptor/). The methods in the scVelo (http://scvelo.org/) (package wrapped by velociraptor) for calculating
RNA velocity and the CellRank package for estimating
state transitions (http://cellrank.org/) will also be briefly described.

Transcript

1. Interoperability between
   Bioconductor and Python
   for scRNA-seq analysis
   Luke Zappia
   @_lazappi_
   

   View Slide

2. What is interoperability?
   “Ability to quickly and easily switch
   between languages/platforms as
   required”
   

   View Slide

3. Why interoperability?
   1. Take advantage of strengths
   2. Make use of existing packages
   3. Avoid unnecessary reimplementation
   

   View Slide

4. Bulk RNA-seq analysis
   

   View Slide

5. scRNA-seq analysis
   Seurat
   CRAN
   

   View Slide

6. Ecosystems
   

   View Slide

7. How?
   

   View Slide

8. {reticulate} {basilisk}
   B
   Python
   environments
   R/Python
   interface
   scRNA-seq
   objects
   Velocity
   analysis
   

   View Slide

9. Disclaimer
   Most (almost all) of this is not my work
   Package Developer
   @GitHub
   Python Alternative
   

   View Slide

10. {reticulate}
    Kevin Ushey
    @kevinushey
    J.J. Allaire
    @jjallaire
    Yuan Tang
    @terrytangyuan
    RStudio
    rstudio.org
    rpy2
    install.packages("reticulate")
    

    View Slide

11. library(reticulate)
    # Set Python environment
    > use_python("/path/to/my/python")
    # use_virtualenv("my_venv")
    # use_condaenv("my_conda_env")
    # Import Python libraries
    > pandas w- import("pandas")
    # Implicitly convert between R and Python
    > pandas$DataFrame(data = list("Col1" = 1:2, "Col2" = 3:4))
    Col1 Col2
    1 1 3
    2 2 4
    # Explicitly convert between R and Python
    > vec w- 1:4
    > vec
    [1] 1 2 3 4
    > py_list w- r_to_py(vec)
    > py_list
    [1, 2, 3, 4]
    > py_to_r(py_list)
    [1] 1 2 3 4
    {reticulate}
    in R
    

    View Slide

12. ```{r}
    # A normal R chunk
    vec w- 1:4
    vec
    ```
    [1] 1 2 3 4
    ```{python}
    # A native Python chunk
    ls = [5, 6, 7, 8]
    ls
    ```
    [1, 2, 3, 4]
    ```{r}
    # Access Python from R
    mean(py$ls)
    ```
    6.5
    ```{python}
    # Access R from Python
    sum(r.vec) / len(r.vec)
    ```
    2.5
    {reticulate}
    in R Markdown
    

    View Slide

13. Conversion
    R Python
    Single-element vector Scalar
    Multi-element vector List
    List of multiple types Tuple
    Named list Dict
    Matrix/Array NumPy array
    data.frame Pandas DataFrame
    Function Python function
    NULL, TRUE, FALSE None, True, False
    

    View Slide

14. Limitations
    Manage Python environment
    Familiarity with Python syntax
    Only supports common data structures
    

    View Slide

15. {basilisk}
    Aaron Lun
    @LTLA
    Image from Ipipipourax via WikiMedia Commons (CC BY-SA 3.0)
    https://commons.wikimedia.org/wiki/File:Basilik_color%C3%A9.jpg
    BiocManagerw:install("basilisk")
    

    View Slide

16. my_env w- basiliskw:BasiliskEnvironment(
    envname = "my_env",
    pkgname = "myPkg",
    packages = c("pandasw=1.1.2", ww.)
    )
    my_py_fun w- function(ww.) {
    pandas w- import("pandas")
    ww.
    return(output)
    }
    my_r_fun w- function(ww.) {
    output w- basiliskw:basiliskRun(
    env = my_env,
    fun = my_py_fun,
    ww.
    )
    }
    library(myPkg)
    output w- my_r_fun(ww.)
    Set up Python (Conda) environment (first
    time)...
    Run my_py_fun() in the environment...
    Return output
    {basilisk}
    1. Define an environment
    2. Create a {reticulate} function
    2. Wrap the function in the environment
    Developer User
    

    View Slide

17. Advantages
    User doesn’t require Python code
    Automatic environment creation
    Different environments/dependencies
    for each package
    

    View Slide

18. {zellkonverter}
    Aaron Lun
    @LTLA
    Luke Zappia
    @lazappi
    anndata2ri
    BiocManagerw:install("zellkonverter")
    

    View Slide

19. {basilisk}
    .h5ad file
    readH5AD()
    AnnData2SCE() SingleCellExperiment ...
    .h5ad file
    AnnData
    {basilisk}
    writeH5AD()
    SCE2AnnData()
    AnnData
    AnnData2SCE() SingleCellExperiment SCE2AnnData()
    AnnData AnnData
    

    View Slide

20. View Slide

21. {anndata}
    Robrecht Cannoodt
    @rcannood
    {sceasy}
    Vladimir Kiselev
    @wikiselev
    Ni Huang
    @nh3
    install.packages("anndata")
    anndata
    remotesw:install_github("sceasy")
    

    View Slide

22. {velociraptor}
    Kevin Rue-Albrecht
    @kevinrue
    Aaron Lun
    @LTLA
    Charlotte Soneson
    @csoneson
    scvelo
    BiocManagerw:install("velociraptor")
    

    View Slide

23. {basilisk}
    SingleCellExperiment
    scvelo()
    AnnData2SCE()
    AnnData
    scv.tl.velocity(...)
    scv.tl.latent_time(...)
    ...
    AnnData
    X
    SingleCellExperiment
    

    View Slide

24. scVelo
    Volker Bergen
    @Volker Bergen
    pip install scvelo
    

    View Slide

25. RNA velocity
    

    View Slide

26. Dynamical RNA velocity
    

    View Slide

27. View Slide

28. View Slide

29. CellRank
    Marius Lange
    @Marius1311
    pip install cellrank
    

    View Slide

30. View Slide

31. View Slide

32. View Slide

33. Pancreas development
    

    View Slide

34. Summary
    Interoperability between Bioconductor and Python is
    already possible
    {zellkonverter} converts between
    SingleCellExperiment and AnnData objects
    scVelo and CellRank for analysis of dynamic processes
    

    View Slide

35. Thanks!
    Luke Zappia
    @_lazappi_
    @lazappi
    lazappi.id.au
    scvelo.org
    cellrank.org
    Theis Lab
    @fabian_theis
    @ICBmunich
    www.comp.bio
    

    View Slide