PHYSICS-GROUNDED SHARED COORDINATES ENABLE BASELINE-INDEPENDENT FINGERPRINTING OF WHOLE-BRAIN DYNAMICS IN AUTISM MOUSE MODELS

Comparing whole-brain dynamics across individuals with neurodevelopmental conditions remains challenging without common references that preserve interpretability and quantify uncertainty—particularly when healthy baselines are unavailable or when comparing systematically different disease subtypes. We present a framework combining population-trained shared latent coordinates with physics-grounded modelling to produce directly comparable, mechanistic fingerprints of brain dynamics at the individual level. Resting-state functional ultrasound data from N=8 mice (seven Cre-lox autism models spanning four subtypes; one control; 54 bilateral regions) were projected into shared low-dimensional spaces using multi-method fusion (SRM, MCCA, Group PCA/ICA) optimised for cross-subject alignment. Binarised latent trajectories were fitted with pairwise maximum-entropy (Ising) models, enabling energy landscape analysis to extract interpretable descriptors: attractor states, barrier distributions, dwell times, and transition structure. We introduce a variance-balanced multi-observable phase-diagram analysis that places each subject onto a Sherrington–Kirkpatrick reference surface by jointly minimising errors across magnetisation, spin-glass order, and susceptibilities—yielding stable optima with bootstrap uncertainty (placement costs 10⁻⁶–10⁻⁴; σ≈0.155–0.320; μ≈−0.013 to +0.031). Results revealed consistent subtype-level patterns in reachability structure and barrier-height distributions, suggesting mechanistic differences in dynamical stability and state-switching propensity rather than coordinate artefacts. The shared-coordinate design enables baseline-independent relative comparisons across heterogeneous cohorts, while interpretable descriptors—barrier spectra, dwell structure, near-criticality indices—support phenotype stratification and hypothesis generation regarding circuit-level substrates of neurodevelopmental conditions. This modality-agnostic pipeline offers a principled approach to characterising atypical brain dynamics when conventional case-control designs are impractical.