||We investigate the spatial distribution of Ly alpha-emitting galaxies (LAEs) at z approximate to 2.67, selected from the NOAO Deep Wide-Field Survey, using two-point statistics and topological diagnostics adopted from network science. We measure the clustering length, r(0) approximate to 4 h(-1) Mpc, and the bias, b(LAE) = 2.2(-0.1)(+0.2). Fitting the clustering with halo occupation distribution (HOD) models results in two disparate possibilities: (1) where the fraction of central galaxies is < 1 per cent in haloes of mass > 10(12) M-circle dot and (2) where the fraction is approximate to 20 per cent. We refer to these two scenarios as the 'Dusty Core Scenario' for Model#1, since most of the central galaxies in massive haloes are dead in Ly alpha emission, and the 'Pristine Core Scenario' for Model#2, since the central galaxies are bright in Ly alpha emission. Traditional two-point statistics cannot distinguish between these disparate models given the current data sets. To overcome this degeneracy, we generate mock catalogues for each HOD model using a high-resolution N-body simulation and adopt a network statistics approach, which provides excellent topological diagnostics for galaxy point distributions. We find three topological anomalies from the spatial distribution of observed LAEs, which are not reproduced by the HOD mocks. We find that Model#2 matches better all network statistics than Model#1, suggesting that the central galaxies in > 10(12) h(-1)M(circle dot) haloes at z approximate to 2.67 need to be less dusty to be bright as LAEs, potentially implying some replenishing channels of pristine gas such as the cold mode accretion.