Synthetic Data Generation for Classifying Electrophysiological and Morpho-Electrophysiological Neurons from Mouse Visual Cortex

Xavier Vasques, Laura Cif, Institut du Neurone, Montferriez sur lez, France Département de Neurochirurgie, Hôpital Gui de Chauliac, Montpellier, France Service de Neurologie, Lausanne University Hospital, Lausanne, Switzerland · 2025

Read Synthetic Data Generation for Classifying Electrophysiological and Morpho-Electrophysiological Neurons from Mouse Visual Cortex by Xavier Vasques, Laura Cif, Institut du Neurone, Montferriez sur lez, France Département de Neurochirurgie, Hôpital Gui de Chauliac, Montpellier, France Service de Neurologie, Lausanne University Hospital, Lausanne, Switzerland (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-06514.txt.

Introduction

Xavier Vasques and Laura Cif test whether fake data can help sort brain cells by type in mouse visual cortex. True data are scarce, and cells vary a lot, so sorting by electric and shape traits is hard. Their 2025 study tries five ways to make extra training data.

They built the tests on a curated set with 48 electric traits and 24 shape traits per cell. A simple method called SMOTE gave the best gains, lifting accuracy by 0.16 for electric types and 0.12 for mixed types. Deep generative models came close only when finely tuned. The team also checked how close each fake set sat to real cell profiles.

Catalog record

CREATORXavier Vasques, Laura Cif, Institut du Neurone, Montferriez sur lez, France Département de Neurochirurgie, Hôpital Gui de Chauliac, Montpellier, France Service de Neurologie, Lausanne University Hospital, Lausanne, Switzerland
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE80 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-06514

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Synthetic Data Generation for Classifying Electrophysiological and Morpho-Electrophysiological Neurons from Mouse Visual Cortex Xavier Vasques, Laura Cif, Institut du Neurone, Montferriez sur lez, France Département de Neurochirurgie, Hôpital Gui de Chauliac, Montpellier, France Service de Neurologie, Lausanne University Hospital, Lausanne, Switzerland Abstract The accurate classification of neuronal cell types is central to decoding brain function, yet remains hindered by data scarcity and cellular heterogeneity. Here, we benchmarked classical and deep generative synthetic data augmentation strategies—including SMOTE, GANs, VAEs, Normalizing Flows, and DDPMs—for supervised classification of both electrophysiological (e-type) and morpho-electrophysiological (mee-type) neuron types from the mouse visual cortex. Using a curated dataset annotated with 48 electrophysiological and 24 morphological features, we established baseline classifiers and introduced synthetic data generated by each method. Our results demonstrate that SMOTE-based augmentation yields the highest classification accuracies (absolute gains of 0.16 for e-types, 0.12 for mee-types), outperforming deep generative models. GANs approached similar performance when hyperparameters and sample sizes were optimized, but were more sensitive to model specification. In addition, we benchmarked synthetic neuron fidelity by comparing mean absolute errors between synthetic and real class profiles against the natural phenotypic variability observed between real neuronal classes. 1 Introduction The precise classification of neuronal cell types is fundamental for unraveling the structural and functional organization of the mammalian brain. High-throughput technologies such as patch-clamp electrophysiology and three-dimensional morphological reconstructions have facilitated profiling of neuronal properties, revealing a rich diversity of cell types in the cortex and beyond [ 44 , 49 , 48 , 19 , 20 ] . Certain neuronal functions have been speculated to manifest exclusively at the circuit level, necessitating the utilization of connectivity properties when available. In other cases, neural activity alterations occur in a pathological or diseased state, and a decline in neural activity may be indicative of activity-dependent recruitment only of specific neuronal subpopulations, crucial for post-injury locomotion [ 28 ] . Structural properties of neurons—including dendritic and axonal morphologies, soma size, and spine density—can be integrated with electrophysiological features such as resting membrane potential, biophysical attributes, firing rate, as well as protein and mRNA composition, to comprehensively represent neuronal physiological and molecular properties [ 15 ] . A critical challenge remains in addressing the substantial cellular heterogeneity present within neural tissue, both in healthy and diseased states. Functional categorization of distinct neuron types remains a pivotal task that demands multidimensional investigation at the level of individual neurons [ 19 ] . Advancements in neuronal classification methodologies have been achieved through a multifaceted approach encompassing morphological [ 41 , 55 ] , molecular, electrophysiological, transcriptomic, and biophysical analyses, thereby enriching the systematic and reproducible understanding of biological structures and functions [ 56 ] . Comprehensive datasets remain scarce, particularly for rare or molecularly distinct subpopulations, and the intrinsic heterogeneity of neuronal phenotypes poses substantial challenges for developing robust and generalizable classification models [ 41 , 55 ] . Imbalanced datasets and insufficient representation of minority classes often limit the accuracy and reliability of machine learning approaches in neuroscience [ 48 ] . The advent of computational neuroscience methodologies has been paramount in quantifying the intricate interplay between neuronal structure and activity [ 22 , 45 ] . A thorough exploration of the morpho-physiological relationships in neurons necessitates extensive modeling to harmonize digitized neuron morphologies with physiological data, a process that demands meticulous manual effort. To accommodate the scarcity of available electrophysiological and morphological data from neurons, multiple data augmentation approaches, pertaining especially to neuron morphologies, have been proposed [ 15 , 31 , 5 , 45 ] ; these approaches use stringent a priori constraints and existing knowledge of neuronal morphology [ 52 ] , which are potential sources of bias. Synthetic data generation has emerged as a promising solution to address these challenges. Classical oversampling techniques like the Synthetic Minority Over-sampling Technique (SMOTE) have been widely used to improve class balance in biological data [ 13 ] . More recently, deep generative models—including Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), Normalizing Flows, and Denoising Diffusion Probabilistic Models (DDPMs)—have demonstrated remarkable success in learning complex, high-dimensional distributions, and generating synthetic data that closely mimic real biological variability [ 40 , 23 , 38 , 29 , 25 ] . In neuroscience, these models have enabled simulation of electrophysiological recordings [ 23 ] , spike trains [ 38 ] , and even high-dimensional morphological and transcriptomic features [ 35 , 20 ] , contributing to improved classifier performance. Despite these advances, the field still lacks standardized and biologically grounded criteria for validating the fidelity of synthetic neuronal data—particularly for morpho-electrophysiological profiles. Most published studies evaluating generative models in neuroscience have focused on visual inspection, distributional similarity of selected features, or improvements in downstream classifier performance as primary measures of synthetic data quality [ 20 , 23 , 38 ] . However, such metrics often overlook the fundamental requirement that synthetic neurons should reproduce not only the central tendencies, but also the natural phenotypic diversity and boundaries characterizing real neuronal populations. Only a handful of recent studies—mainly in transcriptomics and spike train modeling—have begun to emphasize the need for benchmarks that directly compare synthetic-to-real variability against the natural variability between real classes or cell types [ 35 ] . For morpho-electrophysiological neurons in particular, this gap is even more pronounced: while classification frameworks have been established using combined morphological and electrophysiological data [ 19 , 41 ] , comprehensive guidelines or best practices for evaluating synthetic data fidelity in this domain are essentially absent from the literature. Consequently, there remains a need for biologically meaningful benchmarking strategies that can distinguish between synthetic datasets that merely mimic superficial characteristics and those that genuinely recapitulate the full spectrum of natural neuronal diversity. Synthetic neurons can play a pivotal role in the preliminary phases of drug discovery and electroceuticals, enabling initial in silico evaluations on artificial models prior to in vivo implementations and aiding the exploration of neurological and neurodegenerative conditions and potential remedial interventions. In this study, we assessed the effectiveness of both classical and deep generative data augmentation methods—including SMOTE, GANs, VAEs, Normalizing Flows, and DDPMs—for neuronal subtype classification. Leveraging a comprehensive dataset of mouse visual cortex neurons annotated with both electrophysiological and morpho-electrophysiological features [ 19 ] , we established supervised learning baselines and develop data augmentation pipelines that integrate synthetic neurons generated by each method. To provide biologically meaningful validation, we benchmarked synthetic-to-real similarity through classification performance, but also by comparing the variability of synthetic neurons to the natural diversity observed across real neuronal classes. 2 Materials and Methods 2.1 Electrophysiological and Morphological Data Selection The electrophysiological and morphological data used in this research were acquired from the Allen Cell Types Database, accessible at http://celltypes.brain-map.org . The electrophysiological and morphological characteristics of neurons within the mouse visual cortex were defined through an in vitro single-cell characterization platform as presented by Gouwens et al. [ 20 ] , aimed at constructing a standardized dataset. For the current project, 1857 cells exhibiting electrophysiological features and 459 cells with morphological features, originating from the primary visual cortex and proximate higher visual areas, were assessed. Of the electrophysiological cell entities (e-types), 885 were classified as spiny neurons (presumed excitatory), and 972 were categorized as

Read the complete text