dynR

Dynamic connectivity analysis for neurophysiological timeseries in R

R Package
Neuroimaging
Connectivity
Open Source
An R port of the Python dynfc library for computing dynamic connectivity (dynFC) representations — phase-based (LEiDA, Kuramoto) and correlation-based (sliding-window, edge-centric cofluctuation) methods — from multivariate neurophysiological timeseries such as BOLD fMRI, EEG, and LFP.
Published

July 3, 2026

About

dynR computes dynFC representations from preprocessed multivariate timeseries, providing the upstream computation layer for dynamic connectivity analysis. It is a full R port of the Python dynfc library, motivated by reproducibility: R + renv provides a more stable long-term environment than Python dependency chains for research pipelines.

Although the bundled example data and several vignettes use BOLD fMRI, all methods are applicable to any band-limited neurophysiological signal where phase relationships or pairwise correlations carry meaningful information — including EEG, LFP, and MEG.

The outputs of dynR feed directly into stateR for brain state quantification (fractional occupancy, dwell time, Markov transitions).

Pipeline Position

Multivariate timeseries  [N × Tmax]
        │
        ▼
     dynR                 ← this package
  (dynFC representations)
        │
        ▼
     stateR
  (brain state metrics: FO, dwell time, Markov transitions)

Quick Example

library(dynR)

# Simulated timeseries: 80 channels, 300 timepoints
set.seed(42)
ts <- matrix(rnorm(80 * 300), nrow = 80, ncol = 300)

# 1. Bandpass filter (e.g. fMRI: TR = 2 s, 0.01–0.1 Hz)
ts_filt <- apply(ts, 1, bandpass_filter, flp = 0.01, fhi = 0.1, delt = 2)
ts_filt <- t(ts_filt)

# 2. Phase-based: LEiDA + Kuramoto
phases <- hilbert_phases(ts_filt)
dpl    <- dyn_phase_lock(phases)   # dpl$leida: [280 × 80] LEiDA eigenvectors
kop    <- kuramoto(phases)         # kop$metastability, kop$entropy

# 3. Cluster LEiDA vectors → feed into stateR
km <- kmeans(dpl$leida, centers = 5, nstart = 100)

# 4. Correlation-based: sliding-window FC + cofluctuations
sw <- corr_slide(ts_filt, window = 30, step = 5)
ec <- cofluct(ts_filt)             # ec$edge_ts, ec$rss

Features

Function Description
hilbert_phases() Instantaneous phase extraction via the analytic signal
dyn_phase_lock() Dynamic phase-locking matrices (dPL) + LEiDA eigenvectors
get_leida() Leading eigenvector decomposition (LEiDA)
kuramoto() Kuramoto order parameter, metastability, Shannon entropy
corr_slide() Sliding-window Pearson correlation matrices
cofluct() Edge-centric cofluctuation time series + RSS
corr_corr() Correlation-of-correlations (FC recurrence) matrix
bandpass_filter() Zero-phase Butterworth bandpass filter
shannon_entropy() Shannon entropy with optional bit-depth discretisation
dyn_transitions() First-order Markov transition probabilities between brain states

All timeseries inputs follow the [N, Tmax] convention (rows = channels/parcels, columns = timepoints), matching the original Python dynfc package.

Status

dynR is in early development. The API may change without notice, estimation results have not yet been validated against a reference implementation, and the package has not undergone peer review. Use with caution and verify outputs independently before using in any research context.