Fig. 4. Inferring regulatory networks from experimental data.

a GOBI successfully infers predatory interaction from a 30-day abundance time-series data of two unicellular ciliates Paramecium aurelia and Didinium nasutum (data is taken from refs. ^{3, 46}). b GOBI successfully infers the negative feedback loop including negative self-regulations of the synthetic genetic oscillator consisting of a repressor TetR and activator σ²⁸ (data is taken from ref. ⁴⁷). c From time-series data of a three-gene repressilator (data is taken from ref. ⁴⁸), GOBI successfully infers the underlying network. Three direct negative 1D regulations are inferred. Among the three 2D regulations having high TRS, only negative regulations pass the Δ test and surrogate test. d From time series measuring the number of cofactors present at the estrogen-sensitive pS2 promoter after treatment with estradiol (data is taken from ref. ⁵⁰), five 1D regulations have high TRS. However, they are not inferred because they share a common target (dashed box). Among 11 regulations having high TRS, one 2D regulation and two 1D regulations are inferred, passing the Δ test and surrogate test. e From 1000-day time-series data of daily air pollutants and cardiovascular disease occurrence in the city of Hong Kong (data is taken from ref. ²⁰), GOBI finds direct positive causal links from NO₂ and Rspar to the disease. Source data are provided as a Source Data file.