Clustering based method for finding spikes in insect neurons

TL;DR

This paper introduces a clustering-based algorithm for detecting neuronal spikes in invertebrate patch-clamp recordings, especially effective when spike amplitudes are small or obscured by noise.

Contribution

The novel method uses a 3D feature space to distinguish true spikes from noise, improving detection accuracy over traditional thresholding techniques.

Findings

Effective in detecting small amplitude spikes

Separates noise from true spikes using clustering

Open-source implementation available

Abstract

Spikes can be easily detected inmostintracellular recordings as sharp peaks. However, insome experimental preparations,because of unipolar morphology or other characteristicsof the recorded neurons, the sizes of the spikes recorded from the soma can be much smaller. The experimental settings and the quality of the recording can also affect the observed amplitudes of the spikes. Whole-cell patch-clamp recordings from the somata of projection neurons of the antennal lobe in Drosophila or mosquitoes can show spikes with amplitudes as small as 2 mV. Moreover, the observed spikes often ride on relatively large depolarizations, which makes it difficult for the standard thresholding-based approaches to distinguish them from noise or sharp EPSPs present in the signal. For spike detection in such neuronal recordings, we propose a clustering-based algorithm that separates peaks corresponding to action potentials from those corresponding to noise. Candidate peaks, including many noise peaks, are first selected according to their sharpness, and then a feature vector is extracted for each peak. The 3-dimensional feature vector contains the absolute value of the peak voltage, height of the spike, and the magnitude of the second derivative minima attained during the spike. In most recordings, this 3D space reveals two natural clusters, separating the noise peaks from the true action potentials. Some parameters of the algorithm can be optionally altered by the user to improve detection, which comes handy in the few recordings where the default parameters do not work well. In summary, the algorithm facilitates accurate spike detection to enable the interpretation and analysis of patch-clamp data from neuronal recordings in invertebrates. The algorithm is implemented as an freely available open-source tool.