Table 3 Synthetic data: compositional.

From: Edge and modular significance assessment in individual-specific networks

Mult 2 & Pareto—0.7

Method

Median AUC

Mean AUC

MultiLOO-ISN

0.780

0.794

LOO-ISN

0.788

0.788

SSN m

0.758

0.760

KNN log(N),P

0.800

0.800

KNN 5,\(\sqrt{N}\)

0.801

0.803

Optics 5

0.686

0.669

Optics \(\sqrt{N}\)

0.739

0.74

OTS euclidean

0.786

0.786

OTS cosine

0.515

0.519

mOTS cosine

0.542

0.544

mOTS euc

0.800

0.799

mOTS glob

0.739

0.754

Spoutlier -l

0.789

0.787

Cook’s max

0.786

0.793

Cook’s med

0.775

0.776

  1. Bold values indicate the top performer of each column.
  2. Averaged AUC in the context of high heterogeneity and elevate multiplier in synthetic data. KNN methods achieve the best performance, with KNN 5, \(\sqrt{N}\) yielding Mean AUC \(=0.803\) and Median AUC = 0.801, and kNN log(N),P closely following. Euclidean Spoutlier, i.e., Spoutlier-l, OTS euclidean and mOTS euc, Cook’s distance methods, i.e., Cook’s max and Cook’s med, and LOO-ISN are also strong performer, all with Median AUC \(\ge 0.77\). Cosine OTSs methods are not suited for the task and barely better than a random guess.
Back to article page