A trace-based analysis pipeline for coherent and optimized electrophysiological data analysis

Abstract

1The development of large-scale neuronal networks notably relies on the use of point-neuron models to reduce the computational cost of simulations while focusing on integrative neuronal properties. However, the precise tuning of these neuron models remains a major aspect of modeling work to accurately reproduce neuronal properties and understand their implications in network activity. To this end, the precise characterization of neuronal electrophysiological properties, from linear properties to the input-output (I/O) relationship and spike frequency adaptation, from intracellular recordings is a crucial step. Furthermore, the increasing availability of publicly accessible databases opens the possibility of deriving I/O properties for point-neuron models from multiple datasets studying different neuronal populations. However, despite recent advancements in establishing universal data formats for electrophysiological studies, challenges persist due to the absence of standardized protocols (notably for current-clamp experiments) and unified data analysis methods, hindering cross-database comparisons of electrophysiological features. To address these limitations, we developed the TACO pipeline, a free, Python-based pipeline for analyzing databases of current-clamp recordings. The TACO pipeline is designed to be user-friendly, minimizing the need for manual implementation of database-specific data extraction methods and enabling the application of user-defined quality control criteria. The pipeline incorporates robust methods for characterizing neuronal I/O relationships, spike-related feature adaptation, and estimating common experimental artifacts such as bridge errors. These methods have been designed to accommodate variability in database-specific experimental design, the sampling of the input space being of particular importancee. We validated the utility of this approach by demonstrating performance comparable to or exceeding that of machine learning models reported in the literature for neuronal type classification, using protocol-agnostic features extracted by the pipeline. This work highlights the potential of database-independent data analysis tools to enhance cross-database comparability and interoperability, advancing research sustainability and promoting the principles of Open Science.