Connectivity & RSA
Four folders in LaBGAScore wrap external toolboxes, each answering a question the core pipeline does not. All four dependencies are installed separately.
Representational similarity analysis — cosmomvpa/
Built on CoSMoMVPA. RSA asks whether the geometry of neural responses matches a hypothesised structure: are conditions that are conceptually similar also represented similarly in the brain?
Rather than testing whether a region responds more to A than B, RSA compares the full pattern of pairwise dissimilarities against candidate models. This suits questions about how classes of sensation relate to one another — for instance whether visceral and somatic stimulation share a representational structure, or whether interoceptive signals from different organs converge.
Searchlight analyses map representational structure across the brain rather than testing predefined regions.
Decoding accuracy — decoding_toolbox/
Built on The Decoding Toolbox (TDT). Where our PLS-DA and Elastic Net pipelines are built for whole-brain prediction with full inference machinery, TDT is oriented toward classification accuracy maps — particularly searchlight decoding, where a classifier is trained at every location.
Use TDT when the question is where information is present; use the LaBGAScore pipelines when the question is whether a distributed pattern predicts an outcome, and how reliably.
Graph-theoretical connectivity — graphvar/
Generates inputs for GraphVar, which computes graph-theoretical measures on brain connectivity data — degree, clustering coefficient, path length, modularity and related metrics.
Two choices dominate the results and should be made and reported explicitly: the parcellation defining the nodes, and the thresholding applied to the connectivity matrix. Both change the graph metrics substantially.
This pipeline produced the analyses in
proj-IBS-somatization.
Receptor-spatial correlation — juspace/
Built on JuSpace. Tests whether a spatial brain map — a group difference, a multivariate weight map, a PET parametric image — correlates with the distribution of specific neurotransmitter receptors and transporters, derived from independent PET atlases.
This is a way to give a spatial result a neurochemical interpretation: if a pattern tracks serotonergic or dopaminergic receptor density, that constrains what mechanism might produce it. Because brain maps are spatially autocorrelated, the null model matters — spatial autocorrelation-preserving nulls, rather than naive permutation of voxels.
Overlap with CANlab tools. CanlabCore provides atlas objects, region-based extraction and
connectivity utilities, plus the multivariate signatures in
Neuroimaging_Pattern_Masks. For many
questions the CANlab machinery is sufficient and better integrated with the rest of the
pipeline — see the object docs. Reach for these external
toolboxes when you need something they specifically provide.