A multi-class logistic regression algorithm to reliably infer network connectivity from cell membrane potentials

In neuroscience, the structural connectivity matrix of synaptic weights between neurons is one critical factors that determine overall function a network neurons. The mechanisms signal transduction have been intensively studied at different time and spatial scales both cellular molecular levels. While better understanding knowledge some basic processes information handling by has achieved, little known about organization complex neuronal networks. Experimental methods are now available to simultaneously monitor electrical activity large number in real time. analysis data related activities individual can become very valuable tool for study dynamics architecture neural particular, advances optical imaging techniques allow us record up thousands nowadays. However, most efforts focused on calcium signals, lack relevant aspects cell activity. recent years, progresses field genetically encoded voltage indicators shown signals could be well suited spiking events from population Here, we present methodology infer their traces. At first, putative were detected. Then, multi-class logistic regression was used fit penalization term allowed regulate sparseness inferred network. proposed Multi-Class Logistic Regression with L1 (MCLRL) benchmarked against obtained silico simulations. MCLRL properly all tested networks, as indicated Matthew correlation coefficient (MCC). Importantly, accomplished reconstruct among subgroups sampled robustness noise also assessed performances remained high ( MCC >0.95) even extremely conditions (>95% noisy events). Finally, devised procedure optimal regularization term, which allows envision its application experimental data.