Detecting unitary synaptic events with machine learning

Abstract

Spontaneously occurring miniature excitatory postsynaptic currents (mEPSCs) are fundamental electrophysiological events produced by quantal vesicular transmitter release at synapses. Their analysis can provide important information regarding pre- and postsynaptic function. However, the small signal relative to recording noise requires expertise and considerable time for their identification. Furthermore, many mEPSCs smaller than ~8 pA are not well resolved (e.g., those produced at distant synapses or synapses with few receptor channels). Here, we describe an automated approach to detect mEPSCs using a machine learning-based tool. This method, which can be easily generalized to other one-dimensional signals, eliminates inter-observer bias, provides an estimate of its sensitivity and specificity and permits reliable detection of small (e.g., 5 pA) spontaneous unitary synaptic events.