The complete FlowSOM package known from R, now available in Python!
Edited by @BenDeMeurichy to support parallel execution trough XPYSOM.
Please refer to the documentation. In particular, the following resources are available:
You need to have Python 3.9 or newer installed on your system. If you don't have Python installed, we recommend installing Mambaforge.
There are several alternative options to install FlowSOM:
original package:
- Install the latest development version:
- original version:
pip install git+https://github.com/saeyslab/FlowSOM_Python- parallel version:
pip install git+https://github.com/Bendemeurichy/FlowSOM_parallelStarting from an FCS file that is properly transformed, compensated and checked for quality, the following code can be used to run the FlowSOM algorithm:
# Import the FlowSOM package
import flowsom as fs
# Load the FCS file
ff = fs.io.read_FCS("./tests/data/ff.fcs")
# Run the FlowSOM algorithm
fsom = fs.FlowSOM(
ff, cols_to_use=[8, 11, 13, 14, 15, 16, 17], xdim=10, ydim=10, n_clusters=10, seed=42 ,
variant='xpysom', batch=True, batch_size=1000
)
# Plot the FlowSOM results
p = fs.pl.plot_stars(fsom, background_values=fsom.get_cluster_data().obs.metaclustering)
p.show()The FlowSOM class supports multiple implementations of the training algorithm. Currently, the following variants are supported:
- 'original': The original version implemented by the saeyslab.
- 'numba': Implementation using the numbasom at numbasom. This version does not support all the extra parameters but is faster than the original.
- 'xpysom' (default): Implementation using the xpysom library. Performs better, scores quite similarly to the original implementation and supports batches. It also supports all the original parameters.
- 'lr': The original implementation but it uses cosine anealing for learning rate decay instead of linear decay.
- 'batch_som': Batch implementation of the original SOM training function.
Batch training is supported for the 'xpysom' and 'batch_som' variants.
The batch implementation can be enabled by setting the batch parameter to True.
The batch_size parameter can be used to specify the size of the batches if needed, the default for this is #cells // cpu_core_count.
See the changelog.
For questions and help requests or if you found a bug, please use the issue tracker.
If you use FlowSOM in your work, please cite the following papers:
A. Couckuyt, B. Rombaut, Y. Saeys, and S. Van Gassen, “Efficient cytometry analysis with FlowSOM in Python boosts interoperability with other single-cell tools,” Bioinformatics, vol. 40, no. 4, p. btae179, Apr. 2024, doi: 10.1093/bioinformatics/btae179.
S. Van Gassen et al., “FlowSOM: Using self-organizing maps for visualization and interpretation of cytometry data,” Cytometry Part A, vol. 87, no. 7, pp. 636–645, 2015, doi: 10.1002/cyto.a.22625.
M. Manciu, “xpysom: XPySom is a minimalistic implementation of the Self Organizing Maps (SOM).” 2021, doi: xpysom.
N. Marincic, “numbasom: A fast Self-Organizing Map Python library implemented in Numba.,” 2021, doi: numbasom.