OSCR

Dynamics-informed priors (DIP) for neural mass modelling.

Code ↔ Paper

4 matches between paragraphs of the paper and lines of its authors' code, computed by the harvester (lexical-v1). Click a colored paragraph or line to see its counterpart.

The 4 matches
  1. [1] § Methods › Datasets ↔ plot_spectra_params.m, lines 229–268 · score 0.68 · power spectral densities, frequency bins, resolution, post LTG, lamotrigine, levetiracetam
  2. [2] § Methods › Datasets ↔ Spectra/final_spectra_gen.m, lines 3–83 · score 0.66 · post treatment, pre treatment, segmented, drug, baseline, 45 Hz
  3. [3] § Methods › Parameter estimation ↔ Simulations/Cost_lansdcapes/R1_De/DIP_DCM_25_fixed_params/spm_nlsi_GN.m, lines 1–60 · score 0.65 · variational Laplace, free energy, generative models, Gaussian, inversion, log
  4. [4] § Methods › Parameter estimation ↔ Simulations/Cost_lansdcapes/R1_De/DIP_DCM_25_fixed_params/spm_nlsi_GN.m, lines 1–60 · score 0.60 · free energy, generative models, Gaussian, covariance, variational, Bayesian

Paper

Loaded from Europe PMC by your browser, not stored by OSCR: doi.org · Europe PMC

The paper is loaded when this pane is shown.

The authors' code

MATLAB · 613 lines · 19 KB · GPL-3.0 · 2 matches

  1. function [Ep,Cp,Eh,F,L,dFdp,dFdpp] = spm_nlsi_GN(M,U,Y)
  2. % Bayesian inversion of nonlinear models - Gauss-Newton/Variational Laplace
  3. % FORMAT [Ep,Cp,Eh,F] = spm_nlsi_GN(M,U,Y)
  4. %
  5. % [Dynamic] MIMO models
  6. %__________________________________________________________________________
  7. %
  8. % M.IS - function name f(P,M,U) - generative model
  9. % This function specifies the nonlinear model:
  10. % y = Y.y = IS(P,M,U) + X0*P0 + e
  11. % where e ~ N(0,C). For dynamic systems this would be an integration
  12. % scheme (e.g. spm_int). spm_int expects the following:
  13. %
  14. % M.f - f(x,u,P,M)
  15. % M.g - g(x,u,P,M)
  16. % M.h - h(x,u,P,M)
  17. % x - state variables
  18. % u - inputs or causes
  19. % P - free parameters
  20. % M - fixed functional forms and parameters in M
  21. %
  22. % M.FS - function name f(y,M) - feature selection
  23. % This [optional] function performs feature selection assuming the
  24. % generalized model y = FS(y,M) = FS(IS(P,M,U),M) + X0*P0 + e
  25. %
  26. % M.P - starting estimates for model parameters [optional]
  27. %
  28. % M.pE - prior expectation - E{P} of model parameters
  29. % M.pC - prior covariance - Cov{P} of model parameters
  30. %
  31. % M.hE - prior expectation - E{h} of log-precision parameters
  32. % M.hC - prior covariance - Cov{h} of log-precision parameters
  33. %
  34. % U.u - inputs (or just U)
  35. % U.dt - sampling interval
  36. %
  37. % Y.y - outputs (samples x observations x ...)
  38. % Y.dt - sampling interval for outputs
  39. % Y.X0 - confounds or null space (over size(y,1) samples or all vec(y))
  40. % Y.Q - q error precision components (over size(y,1) samples or all vec(y))
  41. %
  42. %
  43. % Parameter estimates
  44. %--------------------------------------------------------------------------
  45. % Ep - (p x 1) conditional expectation E{P|y}
  46. % Cp - (p x p) conditional covariance Cov{P|y}
  47. % Eh - (q x 1) conditional log-precisions E{h|y}
  48. %
  49. % log evidence
  50. %--------------------------------------------------------------------------
  51. % F - [-ve] free energy F = log evidence = p(y|f,g,pE,pC) = p(y|m)
  52. %
  53. %__________________________________________________________________________
  54. % Returns the moments of the posterior p.d.f. of the parameters of a
  55. % nonlinear model specified by IS(P,M,U) under Gaussian assumptions.
  56. % Usually, IS is an integrator of a dynamic MIMO input-state-output model
  57. %
  58. % dx/dt = f(x,u,P)
  59. % y = g(x,u,P) + X0*P0 + e
  60. %
  61. % A static nonlinear observation model with fixed input or causes u
  62. % obtains when x = []. i.e.
  63. %
  64. % y = g([],u,P) + X0*P0e + e
  65. %
  66. % but static nonlinear models are specified more simply using
  67. %
  68. % y = IS(P,M,U) + X0*P0 + e
  69. %
  70. % Priors on the free parameters P are specified in terms of expectation pE
  71. % and covariance pC. The E-Step uses a Fisher-Scoring scheme and a Laplace
  72. % approximation to estimate the conditional expectation and covariance of P
  73. % If the free-energy starts to increase, an abbreviated descent is
  74. % invoked. The M-Step estimates the precision components of e, in terms
  75. % of log-precisions. Although these two steps can be thought of in
  76. % terms of E and N steps they are in fact variational steps of a full
  77. % variational Laplace scheme that accommodates conditional uncertainty
  78. % over both parameters and log precisions (c.f. hyperparameters with hyper
  79. % priors)
  80. %
  81. % An optional feature selection can be specified with parameters M.FS.
  82. %
  83. % For generic aspects of the scheme see:
  84. %
  85. % Friston K, Mattout J, Trujillo-Barreto N, Ashburner J, Penny W.
  86. % Variational free energy and the Laplace approximation.
  87. % NeuroImage. 2007 Jan 1;34(1):220-34.
  88. %
  89. % This scheme handels complex data along the lines originally described in:
  90. %
  91. % Sehpard RJ, Lordan BP, and Grant EH.
  92. % Least squares analysis of complex data with applications to permittivity
  93. % measurements.
  94. % J. Phys. D. Appl. Phys 1970 3:1759-1764.
  95. %
  96. %__________________________________________________________________________
  97. % Copyright (C) 2001-2015 Wellcome Trust Centre for Neuroimaging
  99. % Karl Friston
  100. % $Id: spm_nlsi_GN.m 7279 2018-03-10 21:22:44Z karl $
  102. % options
  103. %--------------------------------------------------------------------------
  104. try, M.nograph; catch, M.nograph = 0; end
  105. try, M.noprint; catch, M.noprint = 0; end
  106. try, M.Nmax; catch, M.Nmax = 128; end
  108. % figure (unless disabled)
  109. %--------------------------------------------------------------------------
  110. if ~M.nograph
  111. Fsi = spm_figure('GetWin','SI');
  112. end
  114. % check integrator
  115. %--------------------------------------------------------------------------
  116. try
  117. M.IS;
  118. catch
  119. M.IS = 'spm_int';
  120. end
  122. % composition of feature selection and prediction (usually an integrator)
  123. %--------------------------------------------------------------------------
  124. try
  125. y = Y.y;
  126. catch
  127. y = Y;
  128. end
  130. try
  132. % try FS(y,M)
  133. %----------------------------------------------------------------------
  134. try
  135. y = feval(M.FS,y,M);
  136. IS = inline([M.FS '(' M.IS '(P,M,U),M)'],'P','M','U');
  138. % try FS(y)
  139. %------------------------------------------------------------------
  140. catch
  141. y = feval(M.FS,y);
  142. IS = inline([M.FS '(' M.IS '(P,M,U))'],'P','M','U');
  143. end
  145. catch
  147. % otherwise FS(y) = y
  148. %----------------------------------------------------------------------
  149. try
  150. IS = inline([M.IS '(P,M,U)'],'P','M','U');
  151. catch
  152. IS = M.IS;
  153. end
  154. end
  156. % converted to function handle
  157. %--------------------------------------------------------------------------
  158. IS = spm_funcheck(IS);
  160. % paramter update eqation
  161. %--------------------------------------------------------------------------
  162. if isfield(M,'f'), M.f = spm_funcheck(M.f); end
  163. if isfield(M,'g'), M.g = spm_funcheck(M.g); end
  164. if isfield(M,'h'), M.h = spm_funcheck(M.h); end
  167. % size of data (samples x response component x response component ...)
  168. %--------------------------------------------------------------------------
  169. if iscell(y)
  170. ns = size(y{1},1);
  171. else
  172. ns = size(y,1);
  173. end
  174. ny = length(spm_vec(y)); % total number of response variables
  175. nr = ny/ns; % number response components
  176. M.ns = ns; % number of samples M.ns
  178. % initial states
  179. %--------------------------------------------------------------------------
  180. try
  181. M.x;
  182. catch
  183. if ~isfield(M,'n'), M.n = 0; end
  184. M.x = sparse(M.n,1);
  185. end
  187. % input
  188. %--------------------------------------------------------------------------
  189. try
  190. U;
  191. catch
  192. U = [];
  193. end
  195. % initial parameters
  196. %--------------------------------------------------------------------------
  197. try
  198. spm_vec(M.P) - spm_vec(M.pE);
  199. fprintf('\nParameter initialisation successful\n')
  200. catch
  201. M.P = M.pE;
  202. end
  204. % time-step
  205. %--------------------------------------------------------------------------
  206. try
  207. dt = Y.dt;
  208. catch
  209. dt = 1;
  210. end
  212. % precision components Q
  213. %--------------------------------------------------------------------------
  214. try
  215. Q = Y.Q;
  216. if isnumeric(Q), Q = {Q}; end
  217. catch
  218. Q = spm_Ce(ns*ones(1,nr));
  219. end
  220. nh = length(Q); % number of precision components
  221. nq = ny/length(Q{1}); % for compact Kronecker form of M-step
  224. % prior moments (assume uninformative priors if not specifed)
  225. %--------------------------------------------------------------------------
  226. pE = M.pE;
  227. try
  228. pC = M.pC;
  229. catch
  230. np = spm_length(M.pE);
  231. pC = speye(np,np)*exp(16);
  232. end
  234. % confounds (if specified)
  235. %--------------------------------------------------------------------------
  236. try
  237. nb = size(Y.X0,1); % number of bins
  238. nx = ny/nb; % number of blocks
  239. dfdu = kron(speye(nx,nx),Y.X0);
  240. catch
  241. dfdu = sparse(ny,0);
  242. end
  243. if isempty(dfdu), dfdu = sparse(ny,0); end
  246. % hyperpriors - expectation (and initialize hyperparameters)
  247. %--------------------------------------------------------------------------
  248. try
  249. hE = M.hE;
  250. if length(hE) ~= nh
  251. hE = hE + sparse(nh,1);
  252. end
  253. catch
  254. hE = sparse(nh,1) - log(var(spm_vec(y))) + 4;
  255. end
  256. h = hE;
  258. % hyperpriors - covariance
  259. %--------------------------------------------------------------------------
  260. try
  261. ihC = spm_inv(M.hC);
  262. if length(ihC) ~= nh
  263. ihC = ihC*speye(nh,nh);
  264. end
  265. catch
  266. ihC = speye(nh,nh)*exp(4);
  267. end
  270. % unpack covariance
  271. %--------------------------------------------------------------------------
  272. if isstruct(pC);
  273. pC = spm_diag(spm_vec(pC));
  274. end
  276. % dimension reduction of parameter space
  277. %--------------------------------------------------------------------------
  278. V = spm_svd(pC,0);
  279. nu = size(dfdu,2); % number of parameters (confounds)
  280. np = size(V,2); % number of parameters (effective)
  281. ip = (1:np)';
  282. iu = (1:nu)' + np;
  284. % second-order moments (in reduced space)
  285. %--------------------------------------------------------------------------
  286. pC = V'*pC*V;
  287. uC = speye(nu,nu)/1e-8;
  288. ipC = inv(spm_cat(spm_diag({pC,uC})));
  290. % initialize conditional density
  291. %--------------------------------------------------------------------------
  292. Eu = spm_pinv(dfdu)*spm_vec(y);
  293. p = [V'*(spm_vec(M.P) - spm_vec(M.pE)); Eu];
  294. Ep = spm_unvec(spm_vec(pE) + V*p(ip),pE);
  297. % EM
  298. %==========================================================================
  299. criterion = [0 0 0 0];
  301. C.F = -Inf; % free energy
  302. v = -4; % log ascent rate
  303. dFdh = zeros(nh,1);
  304. dFdhh = zeros(nh,nh);
  305. for k = 1:M.Nmax
  307. % time
  308. %----------------------------------------------------------------------
  309. tStart = tic;
  311. % E-Step: prediction f, and gradients; dfdp
  312. %======================================================================
  313. try
  315. % gradients
  316. %------------------------------------------------------------------
  317. [dfdp,f] = spm_diff(IS,Ep,M,U,1,{V});
  318. dfdp = reshape(spm_vec(dfdp),ny,np);
  320. % check for stability
  321. %------------------------------------------------------------------
  322. normdfdp = norm(dfdp,'inf');
  323. revert = isnan(normdfdp) || normdfdp > exp(32);
  325. catch
  326. revert = true;
  327. end
  329. if revert && k > 1
  330. for i = 1:4
  332. % reset expansion point and increase regularization
  333. %--------------------------------------------------------------
  334. v = min(v - 2,-4);
  336. % E-Step: update
  337. %--------------------------------------------------------------
  338. p = C.p + spm_dx(dFdpp,dFdp,{v});
  339. Ep = spm_unvec(spm_vec(pE) + V*p(ip),pE);
  341. % try again
  342. %--------------------------------------------------------------
  343. try
  344. [dfdp,f] = spm_diff(IS,Ep,M,U,1,{V});
  345. dfdp = reshape(spm_vec(dfdp),ny,np);
  347. % check for stability
  348. %----------------------------------------------------------
  349. normdfdp = norm(dfdp,'inf');
  350. revert = isnan(normdfdp) || normdfdp > exp(32);
  352. catch
  353. revert = true;
  354. end
  356. % break
  357. %--------------------------------------------------------------
  358. if ~revert, break, end
  360. end
  361. end
  363. % convergence failure
  364. %----------------------------------------------------------------------
  365. if revert
  366. error('SPM:spm_nlsi_GN','Convergence failure.');
  367. end
  370. % prediction error and full gradients
  371. %----------------------------------------------------------------------
  372. e = spm_vec(y) - spm_vec(f) - dfdu*p(iu);
  373. J = -[dfdp dfdu];
  376. % M-step: Fisher scoring scheme to find h = max{F(p,h)}
  377. %======================================================================
  378. for m = 1:8
  380. % precision and conditional covariance
  381. %------------------------------------------------------------------
  382. iS = sparse(0);
  383. for i = 1:nh
  384. iS = iS + Q{i}*(exp(-32) + exp(h(i)));
  385. end
  386. S = spm_inv(iS);
  387. iS = kron(speye(nq),iS);
  388. Pp = real(J'*iS*J);
  389. Cp = spm_inv(Pp + ipC);
  391. % precision operators for M-Step
  392. %------------------------------------------------------------------
  393. for i = 1:nh
  394. P{i} = Q{i}*exp(h(i));
  395. PS{i} = P{i}*S;
  396. P{i} = kron(speye(nq),P{i});
  397. JPJ{i} = real(J'*P{i}*J);
  398. end
  400. % derivatives: dLdh = dL/dh,...
  401. %------------------------------------------------------------------
  402. for i = 1:nh
  403. dFdh(i,1) = trace(PS{i})*nq/2 ...
  404. - real(e'*P{i}*e)/2 ...
  405. - spm_trace(Cp,JPJ{i})/2;
  406. for j = i:nh
  407. dFdhh(i,j) = - spm_trace(PS{i},PS{j})*nq/2;
  408. dFdhh(j,i) = dFdhh(i,j);
  409. end
  410. end
  412. % add hyperpriors
  413. %------------------------------------------------------------------
  414. d = h - hE;
  415. dFdh = dFdh - ihC*d;
  416. dFdhh = dFdhh - ihC;
  417. Ch = spm_inv(real(-dFdhh));
  419. % update ReML estimate
  420. %------------------------------------------------------------------
  421. dh = spm_dx(dFdhh,dFdh,{4});
  422. dh = min(max(dh,-1),1);
  423. h = h + dh;
  425. % convergence
  426. %------------------------------------------------------------------
  427. dF = dFdh'*dh;
  428. if dF < 1e-2, break, end
  430. end
  433. % E-Step with Levenberg-Marquardt regularization
  434. %======================================================================
  436. % objective function: F(p) = log evidence - divergence
  437. %----------------------------------------------------------------------
  438. L(1) = spm_logdet(iS)*nq/2 - real(e'*iS*e)/2 - ny*log(8*atan(1))/2; ...
  439. L(2) = spm_logdet(ipC*Cp)/2 - p'*ipC*p/2;
  440. L(3) = spm_logdet(ihC*Ch)/2 - d'*ihC*d/2;
  441. F = sum(L);
  443. % record increases and reference log-evidence for reporting
  444. %----------------------------------------------------------------------
  445. try
  446. F0;
  447. if ~M.noprint
  448. fprintf(' actual: %.3e (%.2f sec)\n',full(F - C.F),toc(tStart))
  449. end
  450. catch
  451. F0 = F;
  452. end
  454. % if F has increased, update gradients and curvatures for E-Step
  455. %----------------------------------------------------------------------
  456. if F > C.F || k < 3
  458. % accept current estimates
  459. %------------------------------------------------------------------
  460. C.p = p;
  461. C.h = h;
  462. C.F = F;
  463. C.L = L;
  464. C.Cp = Cp;
  466. % E-Step: Conditional update of gradients and curvature
  467. %------------------------------------------------------------------
  468. dFdp = -real(J'*iS*e) - ipC*p;
  469. dFdpp = -real(J'*iS*J) - ipC;
  471. % decrease regularization
  472. %------------------------------------------------------------------
  473. v = min(v + 1/2,4);
  474. str = 'EM:(+)';
  476. else
  478. % reset expansion point
  479. %------------------------------------------------------------------
  480. p = C.p;
  481. h = C.h;
  482. Cp = C.Cp;
  484. % and increase regularization
  485. %------------------------------------------------------------------
  486. v = min(v - 2,-4);
  487. str = 'EM:(-)';
  489. end
  491. % E-Step: update
  492. %======================================================================
  493. dp = spm_dx(dFdpp,dFdp,{v});
  494. p = p + dp;
  495. Ep = spm_unvec(spm_vec(pE) + V*p(ip),pE);
  499. % Graphics
  500. %======================================================================
  501. if exist('Fsi', 'var')
  502. spm_figure('Select', Fsi)
  505. % reshape prediction if necessary
  506. %------------------------------------------------------------------
  507. e = spm_vec(e);
  508. f = spm_vec(f);
  509. try
  510. e = reshape(e,ns,nr);
  511. f = reshape(f,ns,nr);
  512. end
  514. % subplot prediction
  515. %------------------------------------------------------------------
  516. x = (1:size(e,1))*dt;
  517. xLab = 'time (seconds)';
  518. try
  519. if length(M.Hz) == ns
  520. x = M.Hz;
  521. xLab = 'Frequency (Hz)';
  522. end
  523. end
  526. % plot real or complex predictions
  527. %------------------------------------------------------------------
  528. tstr = sprintf('%s: %i','prediction and response: E-Step',k);
  529. if isreal(spm_vec(y))
  531. subplot(2,1,1)
  532. plot(x,real(f)), hold on
  533. plot(x,real(f + e),':'), hold off
  534. xlabel(xLab)
  535. title(tstr,'FontSize',16)
  536. grid on
  538. else
  539. subplot(2,2,1)
  540. plot(x,real(f)), hold on
  541. plot(x,real(f + e),':'), hold off
  542. xlabel(xLab)
  543. ylabel('real')
  544. title(tstr,'FontSize',16)
  545. grid on
  547. subplot(2,2,2)
  548. plot(x,imag(f)), hold on
  549. plot(x,imag(f + e),':'), hold off
  550. xlabel(xLab)
  551. ylabel('imaginary')
  552. title(tstr,'FontSize',16)
  553. grid on
  554. end
  556. % subplot parameters
  557. %--------------------------------------------------------------
  558. subplot(2,1,2)
  559. bar(full(V*p(ip)))
  560. xlabel('parameter')
  561. tstr = 'conditional [minus prior] expectation';
  562. title(tstr,'FontSize',16)
  563. grid on
  564. drawnow
  566. end
  568. % convergence
  569. %----------------------------------------------------------------------
  570. dF = dFdp'*dp;
  571. if ~M.noprint
  572. fprintf('%-6s: %i %6s %-6.3e %6s %.3e ',str,k,'F:',full(C.F - F0),'dF predicted:',full(dF))
  573. end
  574. criterion = [(dF < 1e-1) criterion(1:end - 1)];
  575. if all(criterion)
  576. if ~M.noprint
  577. fprintf(' convergence\n')
  578. end
  579. break
  580. end
  582. % F_all(k)=C.F;
  583. % Ep_all(1,k) = Ep.R(1);
  584. % Ep_all(2,k) = Ep.R(2);
  585. % Ep_all(3,k) = Ep.T(1);
  586. % Ep_all(4,k) = Ep.T(2);
  587. % Ep_all(5,k) = Ep.G;
  588. % Ep_all(6,k) = Ep.H(1);
  589. % Ep_all(7,k) = Ep.H(2);
  590. % Ep_all(8,k) = Ep.H(3);
  591. % Ep_all(9,k) = Ep.H(4);
  592. % Ep_all(10,k) = Ep.H(5);
  593. % Ep_all(11,k) = Ep.A{1};
  594. % Ep_all(12,k) = Ep.A{2};
  595. % Ep_all(13,k) = Ep.A{3};
  596. % Ep_all(14,k) = Ep.D;
  597. % Ep_all(15,k) = Ep.I;
  598. end
  600. if exist('Fsi', 'var')
  601. spm_figure('Focus', Fsi)
  602. end
  604. % outputs
  605. %--------------------------------------------------------------------------
  606. Ep = spm_unvec(spm_vec(pE) + V*C.p(ip),pE);
  607. Cp = V*C.Cp(ip,ip)*V';
  608. Eh = C.h;
  609. F = C.F;
  610. L = C.L;
  612. % F_all=F_all;
  613. % Ep_all=Ep_all;

spm_nlsi_GN.m at commit f00781c, under GPL-3.0 · at the source

Overview

Authors: Alessia Caccamo1,2, Dominic M. Dunstan1,2, Mark P. Richardson3, Alexander D. Shaw4, Marc Goodfellow1,2
ORCID iDs: Alessia Caccamo, Dominic M. Dunstan, Mark P. Richardson, Alexander D. Shaw, Marc Goodfellow
  1. Department of Mathematics and Statistics, University of Exeter, Exeter, United Kingdom
  2. Living Systems Institute, University of Exeter, Exeter, United Kingdom
  3. Department of Basic and Clinical Neuroscience, King’s College London, London, United Kingdom
  4. Department of Psychology, Faculty of Health & Life Sciences, University of Exeter, Exeter, United Kingdom
Institutions: University of Exeter (United Kingdom); King's College London (United Kingdom)
Journal: Imaging neuroscience (Cambridge, Mass.), volume 4, article IMAG.a.1250
Dates: received 30 September 2025; accepted 27 April 2026; published online 29 May 2026
Type: Research article · Language: English
License: CC BY
Identifiers: DOI 10.1162/imag.a.1250 · PMID 42232074 · PMCID PMC13224311 · OpenAlex W4414606096
Open access: diamond, a free copy (OpenAlex)
Preprint: doi:10.1101/2025.09.26.678721
Status: code verified
Categories: EEG (modality), none (in silico) (organism), epilepsy (population)
Methods: Spectral & time-frequency, Preprocessing, Statistics, Single-unit activity, calcium imaging, Machine learning
Keywords: dynamic causal modelling, spectral analysis of M/EEG, variational Bayes, parameter estimation, genetic algorithms, pharmacodynamics of anti-seizure medication
Topic: Neural Networks and Applications (Artificial Intelligence, Computer Science), according to OpenAlex
Funding: EPSRC (UKRI1460, EP/T017856/1, 2407565); Wellcome grant (226709/Z/22/Z)
Citations: not cited yet (Europe PMC); 78 references in the paper

Abstract

Neural mass models (NMMs) are important mathematical tools for inferring hidden neural mechanisms that generate healthy and pathological brain activities. A critical step in the inference process is parameter estimation, which calibrates NMMs based on measured neuroimaging data. While parameter estimation can be conducted via various approaches, one of the most influential methods is dynamic causal modelling (DCM). DCM adopts a Bayesian inference approach that relies on, and is sensitive to, the specification of prior parameter distributions reflecting a priori hypotheses about the causes of data. However, most parameters of NMMs encode neuronal properties that are not directly measurable. For this reason, in the absence of sufficient empirical data and well-founded prior beliefs, inference becomes increasingly susceptible to bias. Therefore, it was imperative to establish a comprehensive strategy for mapping model parameters to data. This study proposes a computational extension of DCM, named DCM with dynamics-informed priors (DIP-DCM), which adopts a genetic algorithm (GA) to map parameter values to model dynamics. Optimal sub-regions of the parameter space were subsequently selected and translated into groups of parameter priors for DCM. DIP-DCM was compared with the “standard” DCM inference and to the standalone GA, using two independent neuroimaging datasets. Results indicated that DIP-DCM models were the best predictors of data and captured key mechanistic signatures of psychiatric disease and pharmacological interventions. Overall, DIP-DCM handled local minima and explored diverse parameter regimes following trajectories informed directly by model dynamics and data. This study suggests that DIP-DCM is an advantageous route to parameter estimation when information is limited, enabling a data-driven derivation of parameter priors in exploratory studies, across different biological contexts and datasets.

Reproduced under the paper's license (CC BY), from the paper cited above.

Repositories

Its files are read in the Code ↔ Paper reader above, with 4 matches between paragraphs and lines of code.

AlessiaCaccamo/Figures_DIP_DCM_25

License: GPL-3.0
State: the link answers, verified on 28 September 2026
Evidence: files inventoried
Commit: f00781c57bf551de87ebf5b17203045a259a276b, 18 March 2026
Languages: MATLAB (60)
Size: 165 files, 60 scripts
Software Heritage: not archived
Found in: “Data and Code Availability”
Holds: README, license file
Not found: CITATION.cff, environment file, tests, continuous integration, documentation
Tools: SPM (14 files), Statistics and Machine Learning Toolbox (6 files), Optimization Toolbox (2 files), Violinplot-Matlab (1 file)
Availability: 1 check, the latest on 28 September 2026: the link answers
  • 28 September 2026: the link answers
62 files
At the source: github.com/AlessiaCaccamo/Figures_DIP_DCM_25.git

AlessiaCaccamo/DIP_DCM_25

License: GPL-3.0
State: the link answers, verified on 28 September 2026
Evidence: files inventoried
Commit: a148b5d5de0891a1dbb200d322809b27f017bf24, 2 October 2025
Languages: MATLAB (40)
Size: 46 files, 40 scripts
Software Heritage: not archived
Found in: “Data and Code Availability”
Holds: README, license file
Not found: CITATION.cff, environment file, tests, continuous integration, documentation
Tools: SPM (8 files), Optimization Toolbox (2 files)
Availability: 1 check, the latest on 28 September 2026: the link answers
  • 28 September 2026: the link answers
42 files
At the source: github.com/AlessiaCaccamo/DIP_DCM_25.git

The paper's code and data availability statement is in the Data section.

Tracing map

Proposed by the machine: these links were found in the paper and verified at the source, without human review. The map will receive a Zenodo DOI once one of the paper's authors has validated it with their ORCID.

What the map holds:

  • 2 repositories of the authors' code, each at its verified commit, with its license and how the link was found in the paper;
  • 100 scripts, each with its path and the digest of its content;
  • 4 matches between paragraphs of the paper and lines of the code (method lexical-v1);
  • neither the text of the paper nor the code itself.

Its JSON (tracing-map.json) is deposited on Zenodo with its DOI once the map is validated.

Data

No dataset and no data link were found in the paper.

Data and Code Availability

All analyses were conducted on data published previously by Biondi et al. (2022) and Shaw et al. (2020). Code used to generate figures including the database of parameter estimates can be found at https://github.com/AlessiaCaccamo/Figures_DIP_DCM_25.git. The DIP-DCM toolbox code is open source, under the terms of the GNU General Public License, and can be found at https://github.com/AlessiaCaccamo/DIP_DCM_25.git. This works on Windows, Linux, and macOS with an installed version of SPM12, and was not tested with other SPM versions. Code includes third-party functions (SPM, https://www.fil.ion.ucl.ac.uk/spm), with their respective copyright.

Reproduced under the paper's license (CC BY), from the paper cited above.

Versions

The history of this record: each version stored by the harvester or made by a correction of its authors or of the maintainers of its code, and what changed in its facts. The texts of the paper (its abstract, its availability statements) are not part of it; versions that changed only those are not listed.

Version 2, 28 September 2026

  • Authors: added Dominic M. Dunstan (0000-0003-1973-1404); Alexander D. Shaw (0000-0001-5741-7526); Marc Goodfellow (0000-0002-7282-7280); removed Dominic M. Dunstan; Alexander D. Shaw; Marc Goodfellow

Version 1, 28 September 2026: the first record

Recorded: type, language, journal, volume, pages, dates, 5 authors, 6 keywords, 2 funders, 76 references.

Cite

This paper

Caccamo, A., Dunstan, D. M., Richardson, M. P., Shaw, A. D., & Goodfellow, M. (2026). Dynamics-informed priors (DIP) for neural mass modelling. Imaging neuroscience (Cambridge, Mass.), 4, IMAG.a.1250. https://doi.org/10.1162/imag.a.1250

BibTeX

@article{caccamo2026dynamics,
author = {Caccamo, Alessia and Dunstan, Dominic M. and Richardson, Mark P. and Shaw, Alexander D. and Goodfellow, Marc},
title = {{Dynamics-informed priors (DIP) for neural mass modelling}},
journal = {Imaging neuroscience (Cambridge, Mass.)},
year = {2026},
month = may,
volume = {4},
pages = {IMAG.a.1250},
publisher = {MIT Press},
issn = {2837-6056},
doi = {10.1162/imag.a.1250},
url = {https://doi.org/10.1162/imag.a.1250},
pmid = {42232074},
pmcid = {PMC13224311}
}

RIS

TY - JOUR
AU - Caccamo, Alessia
AU - Dunstan, Dominic M.
AU - Richardson, Mark P.
AU - Shaw, Alexander D.
AU - Goodfellow, Marc
TI - Dynamics-informed priors (DIP) for neural mass modelling
T2 - Imaging neuroscience (Cambridge, Mass.)
J2 - Imaging Neurosci (Camb)
PY - 2026
DA - 2026/05/29
VL - 4
SP - IMAG.a.1250
SN - 2837-6056
PB - MIT Press
DO - 10.1162/imag.a.1250
UR - https://doi.org/10.1162/imag.a.1250
LA - en
ER -

CSL-JSON

{
"id": "10.1162/imag.a.1250",
"type": "article-journal",
"title": "Dynamics-informed priors (DIP) for neural mass modelling",
"container-title": "Imaging neuroscience (Cambridge, Mass.)",
"author": [
{
"family": "Caccamo",
"given": "Alessia"
},
{
"family": "Dunstan",
"given": "Dominic M."
},
{
"family": "Richardson",
"given": "Mark P."
},
{
"family": "Shaw",
"given": "Alexander D."
},
{
"family": "Goodfellow",
"given": "Marc"
}
],
"container-title-short": "Imaging Neurosci (Camb)",
"volume": "4",
"page": "IMAG.a.1250",
"DOI": "10.1162/imag.a.1250",
"PMID": "42232074",
"PMCID": "PMC13224311",
"ISSN": "2837-6056",
"publisher": "MIT Press",
"URL": "https://doi.org/10.1162/imag.a.1250",
"language": "en",
"issued": {
"date-parts": [
[
2026,
5,
29
]
]
}
}

The tracing map gets a citation of its own once an author has validated it and it has a DOI.

Similar papers

The papers with a page that share the most with this one: the tools found in their code, their categories, datasets, cited references and authors, the rarest counting most.

[1] doi:10.1162/imag.a.1249 [code]
Global search metaheuristics for neural mass model calibration.
Journal: Imaging neuroscience (Cambridge, Mass.)
In common: Optimization Toolbox, Statistics and Machine Learning Toolbox, none (in silico), EEG, 9 references, 3 authors
[2] doi:10.1371/journal.pcbi.1014222 [code]
Neural population models for EEG: From Canonical models to alternative model structures.
Journal: PLoS computational biology
In common: EEG, 6 references
[3] doi:10.1007/s12021-025-09759-w [code]
Limitations of Variational Laplace-Based Dynamic Causal Modelling for Multistable Cortical Circuits.
Journal: Neuroinformatics
In common: SPM, Statistics and Machine Learning Toolbox, 4 references
[4] doi:10.1162/imag.a.1234 [code]
Multiscale parcellation of dynamic causal models of the brain.
Journal: Imaging neuroscience (Cambridge, Mass.)
In common: SPM, Statistics and Machine Learning Toolbox, 3 references
[5] doi:10.1038/s41467-026-76306-9 [code]
Non-invasive characterization of perivascular subarachnoid spaces.
Journal: Nature communications
In common: Violinplot-Matlab, Optimization Toolbox, SPM, 1 other tool
[6] doi:10.1016/j.bbih.2026.101299 [code]
Multimodal approach to identify neuropsychophysiological subgroups in myalgic encephalomyelitis/chronic fatigue syndrome and their relevance for rehabilitation: protocol for a mechanistic cross-sectional and longitudinal study.
Journal: Brain, behavior, & immunity - health
In common: Violinplot-Matlab, Optimization Toolbox, SPM, 1 other tool
[7] doi:10.1038/s41467-026-74565-0 [code]
The functional neurobiology of dispositions towards negative emotions.
Journal: Nature communications
In common: Violinplot-Matlab, Optimization Toolbox, SPM, 1 other tool
[8] doi:10.1002/hbm.70577 [code]
Disgust Propensity, Not Disgust Sensitivity, Shapes the Reactivity of a Subjective Disgust Circuit in Humans.
Journal: Human brain mapping
In common: Violinplot-Matlab, Optimization Toolbox, SPM, 1 other tool
[9] doi:10.1038/s41467-026-71151-2 [code]
Common and distinct neural correlates of social interaction processing and theory of mind in narratives.
Journal: Nature communications
In common: Violinplot-Matlab, Optimization Toolbox, SPM, 1 other tool
[10] doi:10.1371/journal.pcbi.1014022 [code]
Emergence of multifrequency activity in a laminar neural mass model.
Journal: PLoS computational biology
In common: 4 references

Contribute

The authors of this paper can claim it, correct its record and validate its tracing map, and the maintainers of its code (its owner, or a public member of its organization) correct what it says of their repository; anyone signed in can ask for its removal. Every request goes to OSCR's own machine, which answers it; your account page follows them.

Sign in with ORCID to claim this paper as one of its authors, correct its record or validate its tracing map: when the paper's metadata lists your ORCID iD, you are recognized at once. Maintainers of its code: sign in with GitHub, then claim the repository on your account page.

Request its removal

To ask OSCR to remove this record, the copies of its authors' scripts or its tracing map, use the removal request page: signed in, you say who you are, what to remove and why, then review and confirm the request. Published rules decide every request (how).

Discussion, reproductions, activity

Discussion: questions and error reports about this paper and its code, from signed-in readers and its authors. It opens with sign-in.

Reproductions: reports from readers who ran the authors' code: what they reproduced, with which environment, commit and data. It opens with sign-in.

Activity: what happens around this paper: new versions of its record, its map's validation, discussions and reproductions. It opens with sign-in.