forest-fire-clustering

Clustering Method Inspired by Forest Fire Dynamics

Description

Forest Fire Clustering is an efficient and interpretable clustering method for discovering and validating cell types in single-cell sequencing analysis. Different than the existing methods, our clustering algorithm makes minimal prior assumptions about the data and provides point-wise posterior probabilities for internal validation. Additionally, it computes point-wise label entropies that can highlight novel transition cell types de novo along developmental pseudo-time manifolds. Lastly, our inductive algorithm is able to make robust inferences in an online-learning context.

Getting Started

Dependencies

• python >= 3.6
  • numpy
  • scipy
  • scikit-learn
  • numba

Installing

Estimated time: 2 mins

pip install forest-fire-clustering

Executing program

"X" is a [sample x feature] matrix

To generate clustering:

from forest_fire_clustering.forest_fire_clustering import FFC
cluster_obj = FFC()
cluster_obj.preprocess(X)
cluster_obj.fit(fire_temp=100)
cluster_obj.cluster_labels

To validate the results:

cluster_obj.validate()
cluster_obj.entropy()
cluster_obj.pval()

cluster_obj.entropy_list # list of entropies of the data point
cluster_obj.pval_list # list of posterior significance values

Authors

Zhanlin Chen, Jeremy Goldwasser, Philip Tuckman, Jing Zhang, Mark Gerstein