Triku - a feature selection method based on nearest neighbors for single-cell data

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 with 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))

After you have run triku, you need to rerun neighbors using the set of selected features. This doesn’t take much time, though.:

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

tk.tl.triku(adata)

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

In some cases (a high number of features is selected or the dataset is too large) you may need to run PCA first and then neighbors.:

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

tk.tl.triku(adata)

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

This may seem repetitive 🙃, but consider that the first PCA/kNN run is to set the ground for triku to work. Then, with the correct selection of features, PCA/neighbors yield the results with the best features.

Note that for the PCA/neighbors combination you don’t need to specify use_rep="X_triku" because PCA already chooses the selected features from adata.var['highly_variable']. If you run neighbors alone, you must specify use_rep="X_triku" because otherwise it will select adata.X['X_pca'] from the first run!

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.

Cite us!

If you want to learn more about how triku works, you can read our paper 1. Don’t forget to cite it if you find triku useful!

1

Alex M Ascensión, Olga Ibáñez-Solé, Iñaki Inza, Ander Izeta, Marcos J Araúzo-Bravo, Triku: a feature selection method based on nearest neighbors for single-cell data, GigaScience, Volume 11, 2022, giac017, https://doi.org/10.1093/gigascience/giac017