rsDCM

Provides a robust and sparse method for group-level Dynamic Causal Modelling (DCM) of functional magnetic resonance imaging (fMRI) data: Student-t weighting of subjects for robustness, combined with a nonlocal product-moment (pMOM) spike-and-slab prior for sparse selection of group-level effects (<doi:10.48550/arXiv.2609.06379>). The package also provides an R implementation of single-subject DCM for fMRI using variational Laplace inversion (Friston et al., 2003 <doi:10.1016/S1053-8119(03)00202-7>), including the bilinear neural state equation and the Buxton-Friston hemodynamic response model, ported from the 'SPM25' (version 25.01.02) toolbox for 'MATLAB'.

Releases

1

Needs compilation

No

Needs R

>= 4.0.0

Config/roxygen2/version

8.0.0

Config/testthat/edition

3

Date

2026-09-24

LazyData

true

• Functional data analysis
• Penalized regression

• expm
• MASS
• Matrix
• methods
• stats
• utils

Installing it adds

1 package, 145 KB of source

Of those, need compilation

1

• knitr
• R.matlab
• rmarkdown
• testthat
• withr

Lines by kind (6,683)

Lines by language

Languages are counted from file extensions.

Files

94

R functions

64

Median

8.5

Mean

14.19

Highest

77

Median lines per function

30

R functions

64

Calls

110

Unconnected

10

Groups

1

Lines counted

2,243

Lines the tests ran

1,440

Tests passed

Yes

Measured with

covr 3.6.5 on R 4.6.1

Has tests

Yes

Test frameworks

testthat

Test cases

30

Lines of tests

486

Test lines per line of code

0.11

• Introduction to rsDCM

README

Yes, 420 words

Release notes file

Yes

Help pages

54

Lines of documentation

1,787

Commits

–

Open pull requests

0

Tagged releases

0

Latest tagged release

–

Issues closed in

–

Pull requests closed in

–

Pull requests merged

–

Created

14 Sep 2026

Licence file reads as

–

Method, repository practices

Across Kay202

First release.

Robust and sparse group-level DCM

• rsdcm() fits a robust, sparse group-level model to subject-level DCM parameters: Student-t weighting of subjects for robustness to outliers, a nonlocal product-moment (pMOM) spike-and-slab prior for sparse selection of group effects, and ReML-estimated between-subject variance components. It returns the group effects, their posterior inclusion probabilities, the per-subject robustness weights, and the learned spike/slab scales. See <arXiv:2609.06379>.
• rsdcm_fit() is a convenience wrapper that assembles the inputs from a list of dcm_estimate() fits (the posterior Ep/Cp restricted to a chosen parameter field) and a between-subject design.
• narps_dcm: subject-level DCM summaries for 48 subjects (with group/gender/age covariates), derived from the openly shared NARPS dataset. It is the runnable real-data example for rsdcm().
• Progress is reported via message() and is silenceable, matching the rest of the package.

Dynamic Causal Modelling for fMRI (R port of SPM25)

• dcm_estimate(), dcm_nlsi_GN(), dcm_int() ported from SPM25 with optimizations: propagator caching in the bilinear integrator, identity-projection fast path in dcm_diff(), diagonal fast paths in dcm_inv() / dcm_logdet(). Roughly 20x faster end-to-end than a literal port on representative DCMs.
• toy_dcm dataset (three regions, 482 scans) for documentation, smoke-testing, and examples.
• Vignette introduction with a worked end-to-end example.
• Help pages for all exported functions.

Parametric Empirical Bayes (PEB)

• Hierarchical (group-level) modelling of DCM parameters across subjects, ported from SPM25's spm_dcm_peb:
• dcm_peb_prepare() / dcm_peb_run(): prepare and fit a second- or third-level PEB by variational Laplace.
• dcm_peb_of_pebs(): third-level PEB-of-PEBs over a directory of subject PEBs, with a between-subject design.
• dcm_peb_design(): build the group-level design matrix.
• dcm_peb_files() / dcm_peb_load(): locate and load subject PEBs from .rds or MATLAB .mat files (.mat needs the suggested R.matlab package).
• dcm_peb_of_pebs() does not write to disk unless a save_path is supplied.

Naming

Ported SPM routines keep the dcm_ prefix (from spm_), to avoid

colliding with an installed SPM-derived package and to make it obvious

which functions come from here: spm_dcm_estimate becomes dcm_estimate,

spm_nlsi_GN becomes dcm_nlsi_GN, spm_int becomes dcm_int, and so on.

Supported model variants

• This release implements only the deterministic, single-state fMRI DCM (non-linear DCM, via a non-empty d array, is also supported).
• The two-state, stochastic, and spectral / cross-spectral-density (`induced`) variants are not yet implemented. dcm_estimate() stops with an informative error if any of these options is switched on, instead of silently running the deterministic model. They are planned for a future update.

Behavioural changes vs. the MATLAB original

• Progress messages from dcm_nlsi_GN() use message() and can be silenced with suppressMessages().
• dcm_estimate() does not write to disk by default. Pass save = TRUE together with a file path to recover the SPM25 behaviour.
• The previous .spm_env global is replaced by a package-internal environment whose mutable settings are exposed through rsdcm_options().