search

theislab / destiny

R package for single cell and other data analysis using diffusion maps
https://theislab.github.io/destiny/
GNU General Public License v3.0
77 stars 11 forks source link

Diffusion maps compared with scanpy #32

Open zhengyuadalia opened 4 years ago

zhengyuadalia commented 4 years ago

Hi, I have pre-processed my single cell RNA-seq dataset in seurat with its standard pipeline, including normalization, identification of highly variable features and scaling the data, and then I imported the data to desitny and got the result of diffusion map. However, when I tried to imported the counts data of my dataset from seurat to scanpy and pre-processed the data with similar steps (displayed as follows) to get diffusion map, the result was different from that in destiny. Even though I have changed variable features in scanpy to the same features produced by seurat and applied in destiny, the results were still diffetrent. For these two tools are both developed by your lab and implement the same algorithm of diffusion map introduced by Coifman and developed by Haghverdi, I want to know why the diffusion maps are different in them and can you give some advice to make my results more steady?

The preprocessing procedures in scanpy: 图片

The diffusion map in destiny: 图片

The diffusion map in scanpy: 图片

Thank you in advance!

flying-sheep commented 4 years ago

Huh, weird! There’s some differences in the implementation, but I’m fuzzy on what exactly at the moment.

Looks like scanpy creates an outlier on the top left. Maybe you should simply plot DC2 and DC3 in scanpy? Anyway, maybe post this bug in the scanpy tracker, there’s more eyes that see it and I’m just basically one person here!

zhengyuadalia commented 4 years ago

Thanks for your reply! I will ask again in scanpy.