Assessing Cross-Contamination in Spike-Sorted Electrophysiology Data.

Area of Science:

• Computational Neuroscience
• Electrophysiology
• Data Analysis

Background:

• Extracellular electrophysiology enables simultaneous recording of thousands of neurons, demanding advanced spike-sorting computational methods.
• Assessing spike-sorting accuracy, particularly the false discovery rate (FDR), is crucial for reliable neural data interpretation.
• Interspike interval (ISI) violations are commonly used to estimate FDR, but their precise relationship and optimal usage remain unclear.

Purpose of the Study:

• To develop and validate an analytical solution for predicting spike-sorting FDR directly from the ISI violation rate (ISI_v).
• To investigate the nonlinear relationship between ISI_v and FDR, considering factors influencing this association.
• To provide a principled method for researchers to assess spike-sorting accuracy in large-scale electrophysiology datasets.

Main Methods:

• Developed an analytical model to predict FDR based on ISI_v.
• Validated the model using in silico Monte Carlo simulations.
• Applied the model to publicly available extracellular electrophysiology datasets from mouse recordings.

Main Results:

• The relationship between ISI_v and FDR is highly nonlinear, influenced by firing frequency, neuronal activity correlation, and contaminant neuron count.
• Predicted median FDRs in public mouse datasets varied significantly, ranging from 3.1% to 50.0%.
• While predicting FDR for single clusters is challenging due to stochasticity and parameter uncertainty, accurate population-level FDR estimation is feasible.

Conclusions:

• The developed analytical solution provides a robust method for estimating spike-sorting FDR from ISI violations.
• Understanding the nonlinear dependencies is key to accurately interpreting ISI_v as a quality metric.
• This work offers a principled approach to enhance spike-sorting accuracy assessment for the growing field of large-scale neural recordings.