Abstract
Human thought is highly flexible and dynamic, achieved by evolving patterns of brain activity across groups of cells. Neuroscience aims to understand cognition in the brain by analysing these intricate patterns. Here, we argue that this goal is impeded by the time format of our data – clock time. The brain is a system with its own dynamics and regime of time, with no intrinsic concern for the human-invented second. A more appropriate time format is cycles of brain oscillations, which coordinate neural firing and are widely implicated in cognition. These brain dynamics do not obey clock time – they start out of tune with clock time and drift apart even further as oscillations unpredictably slow down, speed up, and undergo abrupt changes. Since oscillations clock cognition, their dynamics should critically inform our analysis. We describe brain time warping as a new method to transform data in accordance with brain dynamics, which sets the time axis to cycles of clocking oscillations (a native unit) rather than milliseconds (a foreign unit). We also introduce the Brain Time Toolbox, a software library that implements brain time warping for electrophysiology data and tests whether it reveals information patterns in line with how the brain uses them.
Competing Interest Statement
The authors have declared no competing interest.
Glossary
- Brain oscillations
- Rhythmic fluctuations of brain activity generated by populations of cells
- Brain time
- Time as sequences of cycles of a coordinating brain oscillation
- Brain time warping
- Algorithm that employs dynamic time warping to transform electrophysiology data in accordance with brain time dynamics
- Brain time toolbox
- Software library that implements brain time warping and tests its effects
- Clock time
- Time as sequences of seconds
- Dynamic time warping (DTW)
- Algorithm that can measure the similarity between signals and minimize their difference
- Frequency
- Number of cycles per time window (typically a second)
- Linear Discriminant Analysis (LDA)
- Machine learning method that maximizes the separability between two classes of data by applying linear transformations to it
- Local Field Potential (LFP)
- The electric potential recorded from extracellular space around cells
- Neural signature
- Brain activity that systematically correlates with, in this context, a cognitive process
- Non-stationarity
- A signal is non-stationary when it undergoes spectral changes over time. We focus on frequency drift, variable starting phases, and phase jumps.
- Periodicity
- Fluctuating patterns of a neural signature
- Phase
- Metric to indicate the specific point in the cycle of an oscillation. Two oscillations are in phase when (for example) their peaks align.
- Temporal Generalization Matrix (TGM)
- Representation of how a classifier trained to separate classes of data on one timepoint performs on other timepoints. When a classifier generalizes, it indicates the neural signature remains stable.
- Warping path
- Representation of how two signals need to be resampled to minimize their difference
- Warping source
- Data structure containing potential coordinating brain oscillations used for brain time warping
