Determine k of consensus clustering

suke18 / FEAST

Feature Selection in scRNA-seq

9 stars 2 forks source link

Determine k of consensus clustering #4

Open jyang635 opened 2 years ago

jyang635 commented 2 years ago

Hi Su,

Thanks for developing FEAST. I am interested in this tool and would like to try it on my own dataset. Since the initial clustering result plays an important role in ranking genes, could you give me some advices on choosing "k" for consensus clustering if I do not know the actual number of cell types (classes) in my dataset?

All the best, Junkai

suke18 commented 2 years ago

Hello Junkai,

Thanks for your interest. I can think of two ways to determine the best k.

The Elbow Method: https://medium.com/analytics-vidhya/how-to-determine-the-optimal-k-for-k-means-708505d204eb.
SC3 predicts the optimal k based on random matrix theory (RMT algorithm). https://www.nature.com/articles/nmeth.4236#Sec2.

Please let me know if you need more.

Best, Kenong

jyang635 commented 2 years ago

Hi Su,

Thank you very much for the quick response. I will take a look at the methods above!

All the best, Junkai