CosTaL: An accurate and scalabel graph-based clustering algorithm for high-dimensional single-cell data analysis

---

Introduction

CosTaL use p-L2knng to map k nearest neighbors and construct a kNN graph, which is later refined using Tanimoto coefficient for a better restoration of the original similarity relationships among the cells inside the high-dimensional single-cell dataset.

CosTaL is still under development in support of all platforms. Right now, the implementation of p-L2knng is only supporting the Linux system using executable, which may generate temproary files as input and output for p-L2knng (remember to get them removed if there is a problem happend during the executation, the name is the automatically-generated UUID as a 32-character lowercase hexadecimal string, unless specificed). The kNN maping on Mac and Windows platforms is temporarily using the brute force mapping borrowed from PhenoGraph.

---

Usages

Install

pip install CosTaL

---

Basic clustering (Tanimoto coefficient refined)

import CosTaL as ct

clusters = ct.clustering(data, method = 'mass', nbr_num = 10)

data: scipy.sparse matrices, numpy.ndarray, pandas.DataFrame

method: {None: 'No preprocessing', 'mass': 'Arcsinh5 transformation', 'flow': 'Arcsinh150 transformation', 'spectral flow': 'Arcsinh6000 transformation', 'scrnaseq': 'Scanpy method, see the code for details'}

nbr_num: k for kNN

Tanimoto coefficient + UMAP connectivity (affinity) doubly-refined clustering

This gives the clustering a slightly better accuracy while requires more time to be executed.

clusters = ct.clustering(data, method = 'mass', nbr_num = 10, local = True)

---

Authors

Yijia Li: [email protected]

Other information

• Liscense

This project is licensed under the [MIT] License - see the LICENSE file for details

• Developemnt status:

Alpha

References

[1] Anastasiu, David C., and George Karypis. "L2knng: Fast exact k-nearest neighbor graph construction with l2-norm pruning." Proceedings of the 24th ACM International on Conference on Information and Knowledge Management. 2015.

[2] McInnes, Leland, John Healy, and James Melville. "Umap: Uniform manifold approximation and projection for dimension reduction." arXiv preprint arXiv:1802.03426 (2018).

[3] Wolf, F. Alexander, Philipp Angerer, and Fabian J. Theis. "SCANPY: large-scale single-cell gene expression data analysis." Genome biology 19.1 (2018): 1-5.