3. Benchmark against centrality baselines

On the 1,200-gene Twist1 subnetwork there is no ground truth, so methods are compared by agreement (table below, all 1,200 genes). NetPert tracks betweenness (rho 0.58–0.59) and most closely driver personalised PageRank (rho 0.79), while raw out-degree agrees least (rho 0.51, top-20 overlap only 2/20).

Method	Spearman vs NetPert	Top-20 overlap	Druggable in top-20
NetPert	—	—	9/20
Betweenness (driver-to-response)	0.58	16/20	9/20
Betweenness (full network)	0.59	17/20	8/20
Out-degree	0.51	2/20	6/20
Driver personalised PageRank	0.79	6/20	6/20

Figure 1. Method agreement on the Twist1 network: NetPert–betweenness rho 0.58–0.59, NetPert–degree 0.51, top-20 overlap 16/20.

Caveat: only NetPert uses the response genes as diffusion targets (subset betweenness uses them as path endpoints). Agreement is not accuracy — the independent checks are DIR status, drug enrichment and the assays.

Agreement supplement ()

**Top-k overlap between methods**
NetPert agrees more with betweenness than with raw degree at every cutoff.

**NetPert weight by gene class**
Intermediates carry the weight; responses score low (excluded from own term).

**NetPert vs BC-full**
Spearman rho=0.59 over 1200 genes.

**BC-sub vs BC-full**
Spearman rho=1.00 over 1200 genes.