Fig. 4: Modular time feature representation revealed by TraceContrast.

a Scheme of neuron subtype classification by TraceContrast. N, T, and D denote the dimensions of neurons, the input Ca²⁺ sequence of each neuron and its output features, respectively. b Bilaterally symmetric and hierarchical modularity emerged from neuron subtype classification. The predefined number of clusters (K) is listed below each corresponding image. c The t-SNE plots of the dimensionality reduction corresponding to b. d Time predictability when sampling only within one specific module (K = 5). Lower dashed line represents the chance level. Data are shown as mean ± SEM (n = 5000 trials). Note that one-module-only sampling results in a marked disruption to time prediction accuracy compared to random sampling in all SCN neurons. Similar results were obtained when K = 2, 3, 4. e, f Ca²⁺ signal amplitude and corresponding variance (e) and average MIC in different modules (f). Data are shown as mean ± SEM (n = 1587, 1936, 1211, 835, and 480 for Modules 1, 2, 3, 4 and 5, respectively). Note that the two attributes showed smooth inter-modular gradients running in opposite directions.