Triku - Feature selection for Single Cell Analysis

Triku (hedgehog in euskera) is a feature selection method prepared for Single Cell Analysis. Triku has been prepared to work with scanpy annData objects directly.

You can visit the notebooks at out notebook repo

How do I use triku?

We suppose you are going to run triku using scanpy. To use triku, simply import it and run it in one line:

import triku as tk

tk.tl.triku(adata)

After that, you can find which features are selected adata.var['highly_variable']. The scores for each gene are located in adata.var['triku_distance']. The higher the score, the better.

If you are using scanpy, you must run triku before running sc.pp.pca and sc.pp.neighbors. We recommend running these commands with the following settings:

import scanpy as sc

sc.pp.pca(adata)
sc.pp.neighbors(adata, metric='cosine', n_neighbors=int(0.5 * len(adata) ** 0.5))

You can run triku with raw or log-transformed count matrices. Scores tend to be better in log-transformed matrices, although the results depend on the dataset.