#35 · Systems biology

Adaptive sampling and graph-based system identification test neural dynamics models

The dominant thread is inference of underlying dynamics from recordings, paired with explicit checks on whether the fitted model can be trusted. A discounted Poisson-Gamma model with Thompson sampling selects electrodes under a fixed channel budget to track non-stationary spontaneous activity in dissociated neuronal networks on HD-MEA. A graph-computational framework with GNN system identification characterises evoked responses in human cortical organoids under repeated stimulation, including depression with repetition. A published study fits power-law and inverse Gaussian renewal point process models to sleep slow oscillations across sleep cycles and scalp EEG channels. What is new is the emphasis on adaptive sampling and model-based inference as routes to validity rather than fit alone. The systems differ in scale and modality, so this reflects a shared methodological emphasis, not a common finding.