# iSTTC: A robust method for accurate estimation of intrinsic neural timescales from single-unit recordings

## TL;DR

The paper introduces iSTTC, a new method for accurately estimating intrinsic neural timescales from single-unit recordings, improving reliability and applicability in various data conditions.

## Contribution

The novel iSTTC method addresses biases and instabilities in existing approaches for estimating intrinsic neural timescales from single-unit activity.

## Key findings

- iSTTC provides more accurate intrinsic timescale estimates across a wide range of data conditions.
- iSTTC works effectively on both unsegmented and epoched data, overcoming limitations of existing methods.
- The method increases the number of neurons suitable for analysis, improving representativeness and robustness.

## Abstract

Intrinsic neural timescales (ITs) are an emerging measure of how neural circuits integrate information over time. ITs are dynamically regulated by behavioral context and cognitive demands, making them suitable for mapping high-level cognitive phenomena onto the underlying neural computations. In particular, IT measurements derived from single-unit activity (SUA) offer fine-grained resolution, critical for mechanistically linking individual neuron dynamics to cognition. However, current methods for estimating ITs from SUA suffer significant biases and instabilities, particularly when applied to sparse, noisy, or epoched neural spike data. Here, we introduce the intrinsic Spike Time Tiling Coefficient (iSTTC), a novel metric specifically developed to address these limitations. Leveraging synthetic and experimental single-unit recordings, we systematically assessed the performance of iSTTC relative to traditional approaches. Our findings demonstrate that iSTTC provides more accurate IT estimates across a wide range of conditions, reducing estimation error especially in challenging yet biologically relevant regimes. Crucially, iSTTC can be applied to both unsegmented and epoched data, overcoming a critical limitation of existing methods. Furthermore, iSTTC substantially relaxes inclusion criteria, increasing the fraction of neurons suitable for analysis and thereby improving the representativeness and robustness of IT measurements. The methodological advances introduced by iSTTC represent a substantial step forward in accurately capturing neural circuit dynamics, ultimately enhancing our ability to link neural mechanisms to cognitive phenomena.

Undestanding how the brain integrates information over different timescales is an important question in neuroscience. Intrinsic neural timescales, derived from the autocorrelation structure of neural activity, provide a window into these integration processes. They vary with behavioral context and cognitive demands, and single-unit recordings offer the most precise way to examine them. However, existing estimation methods can be biased, unstable, or overly restrictive. In this study, we introduce iSTTC, a new method designed to measure intrinsic timescales more accurately and more reliably from single-neuron recordings. Using both simulated and real neural data, we show that iSTTC performs better than commonly used approaches, works well on both continuous and trial-based recordings, and allows many more neurons to be included in the analysis. This means that we can obtain more representative and robust measurements of neural dynamics, even under challenging conditions. By improving how intrinsic timescales are estimated, our method helps pave the way toward a deeper understanding of how neural circuits process information across time.