Downstream analysis on aligned samples has been performed by using the _scanpy_ toolkit in python [[1]](#1).
##### Data concatenation
First, data from the different samples were concatenated by using the function _concatenate()_ from _scanpy_.
##### Filtering out cells and genes
Only cells expressing more than 600 and less than 5000 genes were kept in the analysis, while the other were filtered out. Moreover, cells showing a total number of counts higher than 15000 were filtered out.
Also, regarding the timesteps WP and 5h, only cells expressing at least on of the following genes were kept: GFP, Mef2 and twi. Regarding instead the timestep 9396, only cells expressing at least one gene among GFP and twi were kept.
Cells not matching the conditions above, were filtered out.
Finally, genes expressed in less than 3 cells were excluded from the analysis.
##### Normalization and scaling
Data were normnalized and logarithmized by using the functions _scanpy.pp.normalize_total()_ and _scanpy.pp.log1p()_ respectively, with the default parameters.
Then, total counts and percentage of mitochondrial genes where regressed out by using the function _scanpy.pp.regress_out()_, and data were scaled by using _scanpy.pp.scale()_ with the parameter max_value=10.
##### Principal component analysis, neighborhood graph and UMAP
The principal component analysis was performed on the data through the function _scanpy.tl.pca()_, setting the parameter _svd_solver='arpack'_. Neighborhood graph was computed by using _scanpy.pp.neighboors()_ setting the parameters _n_neighboors=10_ and _n_pcs=40_. Finally, UMAP dimensionality reduction was calculated through _scanpy.tl.umap()_ with standard parameters.
A visual inspection of the data in the UMAP space suggested the presence of batch effect.
##### Batch effect correction
### References
<aid="1">[1]</a>
Wolf, F., Angerer, P. & Theis, F.
SCANPY: large-scale single-cell gene expression data analysis.
