Conference on Cognitive Computational Neuroscience

Sharp peaks near zero in spike-train cross-correlation functions (CCFs) are consistent with the presence of a synaptic connection. Since spiking patterns are shaped by different mechanisms and brain states, we hypothesize that factors such as the cellular intrinsic properties, background oscillations and cellular resonances are reflected in these CCFs and can be extracted from them. Here, we first use dynamical systems and computational modeling to characterize spike-train CCFs in terms of these factors. We then train artificial neural networks that classify ionic conductances and time constants. We identify which of the attributes that describe the CCF’s shape (e.g., secondary peaks located away from the primary sharp peak) are most relevant to allow for an efficient classification. Our results highlight an important link between the subthreshold neuronal and spiking levels that is often overlooked when estimating synaptic strengths from spike train pairs.