Deep Learning Models for Atypical Serotonergic Cells Recognition

TL;DR

This work tackles the challenge of identifying serotonergic neurons when their extracellular spike features deviate from the classic ‘typical’ profile. It proposes a convolutional neural network framework that classifies neurons as serotonergic or non-serotonergic using short 4 ms spike windows, trained on a labeled dataset of fluorescently identified cells and augmented with a large synthetic spike dataset to mitigate noise-induced overfitting. The study demonstrates high accuracy on original data (consensus accuracy ~0.984 with AUC ~0.997) and improved robustness on non-homogeneous data (synthetic model ~0.9375 accuracy; 96.2% TP rate), highlighting the method’s potential for real-time intra-experiment neuron identification. This approach enables rapid, label-free discrimination of neuron types based solely on extracellular spike shapes, supporting broader exploration of serotonergic neuron diversity and their roles in brain function both in vitro and potentially in vivo.

Abstract

The serotonergic system modulates brain processes via functionally distinct subpopulations of neurons with heterogeneous properties, including their electrophysiological activity. In extracellular recordings, serotonergic neurons to be investigated for their functional properties are commonly identified on the basis of "typical" features of their activity, i.e. slow regular firing and relatively long duration of action potentials. Thus, due to the lack of equally robust criteria for discriminating serotonergic neurons with "atypical" features from non-serotonergic cells, the physiological relevance of the diversity of serotonergic neuron activities results largely understudied. We propose deep learning models capable of discriminating typical and atypical serotonergic neurons from non-serotonergic cells with high accuracy. The research utilized electrophysiological in vitro recordings from serotonergic neurons identified by the expression of fluorescent proteins specific to the serotonergic system and non-serotonergic cells. These recordings formed the basis of the training, validation, and testing data for the deep learning models. The study employed convolutional neural networks (CNNs), known for their efficiency in pattern recognition, to classify neurons based on the specific characteristics of their action potentials.

Figures (10)

• Figure 1: Summary of the various steps used to implement the model from recorded signals: from neuronal cell the signal is sampled at 40 kHz and recorded as .abf file, then the all events are selected and sent to 10 neural networks with the above architecture for classification (only difference between the architectures is the value of the 2D convolutional kernel with ranges between 20 to 30).
• Figure 2: Example on how single events were isolated and selected. The image depicts the recording of the serotonergic cell A140313#073 and the 4 ms event of triggered at point 1007585, i.e. at 25.189 sec.
• Figure 3: Examples of noise masks collected from the recordings of cell A140724#065 (on the left) and A160127#015 (on the right).
• Figure 4: Example of 4 synthetic action potentials generated by the event triggered at 1007585, i.e. 25.189 sec, of the serotonergic cell A140313#073. Top trace: the original recording of the event. The panels report four action potential obtained by processing the original trace with different noise masks (see methods).
• Figure 5: Examples of action potentials recorded from serotonergic and non-serotonergic neurons in slices of dorsal raphe nucleus. A. Fluorescent protein-labelled (serotonergic) neurons: a1,a2: typical action potentials of serotonergic neurons; note the long interval between spike upstroke and downstroke (UDI) highlighted by the shaded area in all traces. a3-a4: recordings from serotonergic neurons displaying spikes of shorter duration. B. Fluorescent protein-unlabelled (non-serotonergic) neurons: b1: typical biphasic spike of short duration from a non-serotonergic neuron; b2-b4: spikes of variable shape recorded from non-serotonergic neurons. Shaded areas indicate the width of the spike measured by UDI (see methods). Note the overlap in action potential width of some serotonergic and non-serotonergic neurons. Traces are averages of 15-50 sweeps. Calibrations 25 pA (polarity inverted); 1 ms.