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

Probing the architecture of neuronal circuits and the principles that underlie their functional organization remains an important challenge of modern neurosciences. This holds true, in particular, for the inference of neuronal connectivity from large-scale extracellular recordings. Despite the popularity of this approach and a number of elaborate methods to reconstruct networks, the degree to which synaptic connections can be reconstructed from spike-train recordings alone remains controversial. Here, we provide a framework to probe and compare connectivity inference algorithms, using a combination of synthetic and empirical ground-truth data sets, obtained from simulations and parallel single-cell patch-clamp and high-density microelectrode array (HD-MEA) recordings in vitro. We find that reconstruction performance critically depends on the regularity of the recorded spontaneous activity, i.e., their dynamical regime, the type of connectivity, and the amount of available spike train data. We find gross differences between different algorithms, and many algorithms have difficulties in detecting inhibitory connections. We therefore introduce an ensemble artificial neural network (eANN) to improve connectivity inference. We train the eANN on the validated outputs of six established inference algorithms, and show how it improves network reconstruction accuracy and robustness. Overall, the eANN was robust across different dynamical regimes, with shorter recording time, and ameliorated the identification of synaptic connections, in particular inhibitory ones. Results indicated that the eANN also improved the topological characterization of neuronal networks. The presented methodology contributes to advancing the performance of inference algorithms and facilitates our understanding of how neuronal activity relates to synaptic connectivity. Author summaryThis study introduces an ensemble artificial neural network (eANN) to infer neuronal connectivity from multi-unit spike time recordings. We compare the eANN to previous algorithms and validate it using simulations and HD-MEA/patch-clamp datasets. The latter is obtained from three single-cell patch-clamp recordings and high-density microelectrode array (HD-MEA) measurements, in parallel. Our results demonstrate that the eANN outperforms all other algorithms across different dynamical regimes and provides a more accurate description of the underlying topological organization of the studied networks. We also provide a SHAP analysis of the trained eANN to understand which input features of the eANN contribute most to this superior performance. The eANN is a promising approach to improve connectivity inference from spike-train data.