A Multitaper, Causal Decomposition for Stochastic, Multivariate Time Series: Application to High-Frequency Calcium...

Area of Science:

• Computational Neuroscience
• Time Series Analysis
• Data Science

Background:

• Neural data analysis increasingly utilizes causal information to understand circuit connectivity.
• Dimensional reduction is fundamental for analyzing large, multivariate time series data.
• Standard methods often rely solely on zero-lag information, potentially overlooking dynamics.

Purpose of the Study:

• To present a novel multitaper-based decomposition for stochastic, multivariate time series.
• To analyze neural data by incorporating covariance at all time lags (τ).
• To highlight the limitations of methods that only use zero-lag information.

Main Methods:

• Developed a multitaper decomposition method.
• Applied the method to the covariance of time series at all lags, C(τ).
• Compared results with standard methods using only zero-lag information, X(t).

Main Results:

• Demonstrated the effectiveness of the new method in simulated data.
• Validated the approach using neural imaging data.
• Showed that neglecting full causal structure leads to loss of dynamical information.

Conclusions:

• The proposed multitaper decomposition effectively captures causal structure in neural time series.
• Incorporating all time lags provides a more complete dynamical picture than zero-lag methods.
• This approach enhances the analysis of neural circuit connectivity and dynamics.