search

Vivianstats / scImpute

Accurate and robust imputation of scRNA-seq data
https://www.nature.com/articles/s41467-018-03405-7
90 stars 34 forks source link

Parameter setting in tSNE using PBMC datasets #15

Closed Natsu01 closed 6 years ago

Natsu01 commented 6 years ago

Dear Vivian

I'm interested in scImpute and trying to reproduce the tSNE plot using the PBMC data presented in Fig.6 in your paper. I generated the PBMC datasets which consists of 10 cell types each of 500 cells, which are randomly selected and plotted the tSNE results in Python. Here is the results of tSNE.

tsne_pbmc_imputed_10k

In this experiment, I used the following two parameters in scImpute, Kcluster=10 and drop_thre=0.5, and also used the following three parameters in tSNE, perplexity = 20, n_iter = 5000, random_state = 0 . However, when using tSNE, PBMC cells cannot be clearly classified after imputation.

Would you tell me how to set the parameters in scImpute and tSNE to reproduce the PBMC plot ?

Thanks, Natsu

Vivianstats commented 6 years ago

Hello Natsu,

I used the default parameters in tSNE and drop_thre = 0.5, Kcluster = 11 in scImpute. Before applying tSNE, I log-transformed the data to reduce the effect of extreme values. Based on the axes in your plot, it looks like that you were using the original scale? I guess that's why we get different results.

Best, Vivian

Natsu01 commented 6 years ago

Hello Vivian,

Thank you for your reply. Yes, I used the original scale, when applying tSNE. But I could reproduce the tSNE plot like Fig.6 when I used the parameter, Kcluster = 11 in scImpute and the log-transformed data.

Thank you very much.

Best, Natsu

Vivianstats commented 6 years ago

Great! I'll close the issue, but please feel free to let me know if you have any other questions.