Ensemble learning and ground-truth validation of synaptic connectivity inferred from spike trains

• DOI: 10.1371/journal.pcbi.1011964
• Publication Date: 2024-04-29T17:43:40Z
• Authors: Christian Donner; Julian Bartram; Philipp Hornauer; Taehoon Kim; Damian Roqueiro; Andreas Hierlemann; Guillaume Obozinski; Manuel Schröter
• Keywords: Spike train; Ground truth; Biological neural network; Robustness; Multielectrode array

Probing the architecture of neuronal circuits and principles that underlie their functional organization remains an important challenge modern neurosciences. This holds true, in particular, for inference connectivity from large-scale extracellular recordings. Despite popularity this approach a number elaborate methods to reconstruct networks, degree which synaptic connections can be reconstructed spike-train recordings alone controversial. Here, we provide framework probe compare algorithms, using combination synthetic ground-truth vitro data sets, where labels were obtained simultaneous high-density microelectrode array (HD-MEA) patch-clamp We find reconstruction performance critically depends on regularity recorded spontaneous activity, i.e., dynamical regime, type connectivity, amount available data. therefore introduce ensemble artificial neural network (eANN) improve inference. train eANN validated outputs six established algorithms show how it improves accuracy robustness. Overall, demonstrated strong across different regimes, worked well smaller datasets, improved detection especially inhibitory connections. Results indicated also topological characterization networks. The presented methodology contributes advancing facilitates our understanding activity relates connectivity.