Table 3 Comparison between protein complexes defined by small-scale experiments versus those inferred from high-throughput interaction datasets.

From: Identifying robust communities and multi-community nodes by combining top-down and bottom-up approaches to clustering

Ground truth definition source

Network dataset

SpeakEasy output type

NMI

Omega

Precision

Recall

F-score

Precision (overlapping)

Recall (overlapping)

F-score (overlapping)

CYC2008

Collins et al.

Disjoint

0.7237

0.6382

0.9844

0.8259

0.8982

NA

NA

NA

CYC2008

Collins et al.

Overlapping

0.7120

0.5961

0.9845

0.8055

0.8860

0.2151

0.1170

0.1515

CYC2008

Gavin et al.

Disjoint

0.4502

0.4530

0.8915

0.5395

0.6722

NA

NA

NA

CYC2008

Gavin et al.

Overlapping

0.4498

0.4265

0.8837

0.5292

0.6620

0.2105

0.0958

0.1317

MIPS

Collins et al.

Disjoint

0.6669

0.3740

0.9208

0.8701

0.8947

NA

NA

NA

MIPS

Collins et al.

Overlapping

0.6665

0.3880

0.9118

0.8588

0.8845

0.7821

0.1227

0.2122

MIPS

Gavin et al.

Disjoint

0.5155

0.2001

0.8889

0.7238

0.7979

NA

NA

NA

MIPS

Gavin et al.

Overlapping

0.4929

0.2259

0.9053

0.7127

0.7975

0.7143

0.2092

0.3236

SGD

Collins et al.

Disjoint

0.7147

0.5652

0.9597

0.7510

0.8426

NA

NA

NA

SGD

Collins et al.

Overlapping

0.7058

0.5106

0.9733

0.7470

0.8453

0.4766

0.2048

0.2865

SGD

Gavin et al.

Disjoint

0.5474

0.5215

0.9907

0.6255

0.7668

NA

NA

NA

SGD

Gavin et al.

Overlapping

0.5460

0.5130

0.9722

0.6175

0.7553

0.3659

0.0652

0.1107

  1. Table values consist of normalized mutual information (NMI) between predicted and canonical protein complexes.