Detecting unitary synaptic events with machine learning - Automated detection of small spontaneous synaptic events

Aug 3, 2026

Upload by Maeve

Spontaneously occurring miniature excitatory postsynaptic currents (mEPSCs) reflect quantal vesicular transmitter release and are widely used to infer pre- and postsynaptic function. Their identification is hindered by a small signal-to-noise ratio, time-consuming human expertise, and unresolved events below detection thresholds. The work presents an automated machine learning approach that generalizes to other one-dimensional signals, reduces inter-observer bias, and quantifies sensitivity and specificity to enable reliable detection of small spontaneous unitary synaptic events.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages6

Reading Time15 min

AccessFree to download

Background

Limitations of manual detection

Methods

ANN training and testing

Results

Automated detection performance

Significance

Sensitivity, specificity, and reliable event detection

UC San Diego UC San Diego Previously Published Works Title Detecting unitary synaptic events with machine learning. Permalink [https://escholarship.org/uc/item/45g5b3qg](https://escholarship.org/uc/item/45g5b3qg) Journal Proceedings of the National Academy of Sciences, 121(6) Authors Wang, Nien-Shao Marino, Marc Malinow, Roberto Publication Date 2024-02-06 DOI 10.1073/pnas.2315804121 Peer reviewed [eScholarship.org](eScholarship.org) Powered by the California Digital Library University of California RESEARCH ARTICLE BIOPHYSICS AND COMPUTATIONAL BIOLOGY NEUROSCIENCE Detecting unitary synaptic events with machine learning Nien-Shao Wanga, Marc Marinoa, and Roberto Malinowa,1 Contributed by Roberto Malinow; received September 11, 2023; accepted December 14, 2023; reviewed by Bo Li and Simon Rumpel 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 mEPSCsusing 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. synapse | miniature | machine learning The first spontaneously occurring miniature excitatory postsynaptic signals were recorded at the neuromuscular junction (1). These events were subsequently shown to be due to the release of a single vesicle containing neurotransmitter and provided the basis for the quantal theory of synaptic transmission (2, 3) . Since then miniature quantal events have been recorded in numerous preparations, and their properties (e.g., quantal size, quantal content and quantal event frequency) have been used to make conclusions regarding pre- and postsynaptic function (reviewed in ref. 4). For instance, the number or function of receptors is reflected in quantal amplitude; while the number of synapses or their probability of transmitter release correlates with quantal frequency. While there are notable exceptions to such correlations (e.g., refs. 5–7; reviewed in ref. 8), the amplitude and frequency of quantal events continue to be widely used as measures of synaptic function. One important caveat to such use of miniature events is that many quantal events are small and cannot be distinguished from the recording noise. Because of this complication, signaling events that are smaller than the recording noise are generally not detected and thus not counted. An increase in receptor number or receptor function at synapses producing such events can render quantal events greater than the noise and thus detectable upon analysis. Such analyses can misattribute actual postsynaptic changes to be of presynaptic origin [e.g., during long-term potentiation (5–7)] . While some algorithms have been written to detect miniature events (9, 10) in practice many (and of greatest concern, an unknowable number of ) small events are not detected. Furthermore, most algorithm-based detection is time-consuming and requires the eye of a well-trained scientist to confirm or reject detected events. A fast, sensitive, and accurate algorithm that detects and measures miniature events, which does not require human observation, and importantly provides an estimate of the sensitivity and specificity of detecting events of a given amplitude, would be of considerable benefit to the cellular neuroscience community. Artifi

FAQs

Why are miniature excitatory postsynaptic currents (mEPSCs) difficult to detect?

Their amplitude is small relative to recording noise, making them hard to distinguish and requiring extensive expertise and time for manual identification.

What problem does the machine learning approach address?

It detects small signals embedded in noise using a trained ANN, eliminating inter-observer bias and enabling measurement of sensitivity and specificity.

How can the method benefit studies of synaptic function?

By allowing reliable detection of small spontaneous unitary synaptic events (e.g., around 5 pA), it supports more consistent quantification of synaptic changes and reduces misattribution caused by undetected events.

Aug 3, 2026

Upload by Maeve