Table 1 Optimal number of clusters (\(k_{pred}\)) inferred from cell cycle dataset (\(k_{true}\)=5) with a given range of [1, 20] and clustering performance evaluation with inferred \(k_{pred}\).

From: Inferring transcriptomic cell states and transitions only from time series transcriptome data

Algorithm

Gap statistics using GP

Gap statistics using ED

\(k_{pred}\)

F1

ARI

Sil.

\(k_{pred}\)

F1

ARI

Sil.

TRACS (KM)

7

0.536

0.410

0.216

18

0.239

0.158

0.100

TRACS (AC)

7

0.559

0.428

0.218

17

0.353

0.260

0.130

BHC

40

0.119

0.065

− 0.156

    

DPGP

48

0.209

0.148

0.087

    

STEM

–

–

–

–

    

Algorithm

\(k_{given}\)

F1

ARI

Sil.

K-shape

5

0.578

0.442

0.261

GPclust

5

0.410

0.236

0.123

ClusterNet

5

0.280

0.081

− 0.021

  1. As K-shape, GPclust and ClusterNet do not predict the number of clusters, the true number of clusters is given (\(k_{given} = k_{true}\)). (F1 F1 score, ARI adjusted Rand index, Sil. Silhouette score, KM K-means clustering, AC agglomerative clustering, KS K-shape).
Back to article page