THE ENTANGLED LOOP: CODE AND DATA

CODE AND DATA

The trilogy is intended to be replicated and built on, not taken on trust. This page collects all the computational resources used in the three papers, with documentation, and the access procedures for the empirical datasets. We welcome independent replication and will support it actively.

Computational pipelines

The Växjö Interference Connectivity (VIC) pipeline. The complete code used to compute VIC in the canary in the mind paper, including the whole-brain modelling step, the spectral decomposition and the connectome-based predictive modelling. Available at [GitHub link to be inserted on upload day]. The repository includes documentation, example data and a single-command script that runs the full pipeline end to end.

The LSD perturbation analysis. The code used to test the central identity of the One Operator paper, including the construction of the perturbed operator from the 5-HT2A receptor density map, the simultaneous prediction of the turbulence and harmonic-energy shifts, and the optimisation of the single coupling parameter. Available from the authors.

The whole-brain modelling framework. The broader generative modelling code base that underpins both the LSD analysis and the canary in the mind paper, including Stuart-Landau oscillator dynamics, generative effective connectivity and the model fitting procedures. For a comprehensive treatment of the methodology, see the open-access book Whole-brain modelling: Cartography of the dynamics of mind by Deco and Kringelbach (2025).

Datasets

The HCP BANDA cohort. The Boston Adolescent Neuroimaging of Depression and Anxiety dataset used in the canary in the mind paper. The 150 adolescents aged 14 to 17 used in the analysis are the subset of the BANDA 1.1 release of 215 participants who had baseline resting-state fMRI, RCADS scores at both baseline and one-year follow-up, and low-motion data at baseline. Available through the Human Connectome BANDA Project with standard data-access procedures.

The LSD perturbation dataset. The pharmacological resting-state functional MRI dataset used in the One Operator paper. Originally collected by Carhart-Harris and colleagues and made available through Imperial College London. Access procedures are detailed in the One Operator paper.

The 5-HT2A receptor density map. Drawn from the openly available neurotransmitter density in the Cimbi database collected from 210 healthy participants. Available from their public repository.

The HCP cognition cohort. Used in the canary paper for the demonstration that the VIC signature also tracks cognitive variation in healthy adults. Approximately 1,000 participants from the Human Connectome Project main release.

Replication

We invite independent replication of the canary result and of the LSD test of the central identity. Both tests are within reach of any group with access to functional MRI data and the relevant whole-brain modelling expertise.

The updates page will be maintained as a public log of replication attempts and their outcomes, whether successful or not. We will list any replication attempt that we are aware of, and link to the original report wherever possible. We will not curate the log to remove negative replications.

If you are planning a replication and would like our support, please get in touch through the contact page. We are happy to share unpublished documentation, point to relevant prior work, and discuss methodological choices in advance. We are also happy to be listed as collaborators on replication efforts where this would be useful.

Licensing

The computational pipelines are released under an open licence permitting academic and commercial use, with attribution to the trilogy papers. The exact licence text is included in each repository.

The empirical datasets are not ours to license and are governed by the data-sharing agreements of their original consortia, primarily the Human Connectome Project. Please follow those agreements when reusing the data.

A note on computational reproducibility

The whole-brain modelling pipelines are computationally intensive, with model fitting typically taking hours to days on a single workstation depending on the cohort size. We have made every effort to ensure that the pipelines are deterministic given a fixed random seed, but small numerical differences across hardware and software configurations are inevitable in fits of this complexity. If you obtain results that are qualitatively similar to ours but quantitatively different at the third decimal place, this is expected. If you obtain qualitatively different results, please get in touch so that we can work out together where the difference arises.