Comparing Surrogates to Evaluate Precisely Timed Higher-Order Spike Correlations.

Area of Science:

• Computational Neuroscience
• Data Analysis
• Statistical Modeling

Background:

• Surrogate data generation is crucial for hypothesis testing when analytical methods are insufficient, particularly for assessing spike train correlations.
• A key challenge is creating appropriate null-hypothesis distributions without altering single spike train statistics.
• Uniform dithering (UD) is a classical surrogate technique that locally displaces spikes to disrupt fine-timescale temporal properties.

Purpose of the Study:

• To compare the performance of uniform dithering (UD) against five other surrogate techniques for detecting significant spatiotemporal spike patterns.
• To evaluate surrogate performance on both stationary and nonstationary artificial spike train data.
• To identify which statistical features of spike trains are modified by these methods and to what extent.

Main Methods:

• Comparison of UD with five alternative surrogate techniques.
• Evaluation using point process models with constant firing rates.
• Assessment on modeled nonstationary artificial data to detect false positives (FP) in realistic scenarios.

Main Results:

• Uniform dithering (UD) proves inadequate due to spike loss during binning and clipping, leading to a high rate of false positive (FP) pattern detection.
• Alternative surrogate techniques demonstrate superior performance in identifying precisely timed higher-order correlations.
• Analysis of experimental data from macaque monkey pre-/motor cortex during a reaching-and-grasping task.

Conclusions:

• Uniform dithering (UD) is not suitable for analyzing spike train correlations due to its tendency to generate false positives.
• Other tested surrogate methods are more effective for detecting significant spatiotemporal patterns and higher-order correlations in neural data.
• Findings inform the selection of appropriate surrogate techniques for analyzing complex neural recordings.