scvoting-vjbaskar 0.0.2

Projection of on scRNAseq on another and voting. Sam Watcham's thesis chapter 3

scvoting

This is a voting method to obtain the density (smoothed votes) of a 10 sc-RNASeq dataset on a common reference sc-RNASeq data's landscape which can be viewed using umap, fdg etc

The method works with scanpy annData objects only and contains two parts.

Projecting onto reference landscape

This package follows the Xiaonan Wang's method of projecting a 10x sc-RNASeq dataset onto a reference sc-RNASeq data. After projection

Voting and smoothing

• For every cell in the dataset its NN in the reference dataset gets 1 vote.
• Smooth out the votes for each cell in the reference dataset by sharing them equally among 100 NN in it. So if its vote is 40 you share 40/100 among each of them.
• Compute this for WT and mutant datasets.
• Plot their difference or logFC etc on the reference landscape (UMAP)

Example run

from scvoting import Projector
# Initantiation
votp = Projector(adata = adata, ref_adata=adata_niki, ref_hvg=niki_hvg, npcs=25)
# Proj onto ref
votp.project_on_ref()
# Plot PCAs (works only with X11, use jupyter on remote servers)
votp.plotPCA()
# Compute distances between cells in two datasets in PCA space
votp.pwdist()
# Voting starts
votp.voting(tot_votes = 500000)
# Compute distances between cells in reference datasets
votp.pwdist(rtype = 'ref')
# Smooth the voting for better visualisation
votp.vote_smoothing()
# Smoothed votes
results = votp.votes_smooth