AsymmetryNorm

Description of the Trained Models The repository contains a set of hierarchical Bayesian normative models trained separately for each regional brain asymmetry index. Each model estimates how the asymmetry of a specific region depends on age, sex, and dataset membership, while also learning how the variability of the asymmetry changes across the lifespan. The mean function of each model is represented using a B spline basis over age. This allows the model to flexibly capture nonlinear developmental and ageing related changes. The mean also includes an effect of sex and dataset specific adjustments that account for systematic differences between contributing cohorts. The variance of the asymmetry is modeled with a second B spline basis applied to the logarithm of the standard deviation, which allows the model to express age dependent heteroskedasticity rather than assuming constant noise. All parameters are estimated using variational inference with posterior sampling to obtain full uncertainty distributions. The resulting models provide posterior samples for all parameters and are fully self contained and reproducible because every model includes the exact knots used to generate the spline bases. Contents of the output files NetCDF file (.nc)Contains the full posterior for a trained model. This includes sampled values for all parameters such as spline weights for the mean, spline weights for the variance, dataset effects, sex effects, and derived quantities. The file is stored in the standard InferenceData format from ArviZ, which supports diagnostics, summaries, and direct use during prediction. JSON metadata fileContains the metadata required to reconstruct the exact design matrices used during training. This includes the region name, the list of dataset levels and their encoding, the degrees of freedom for both spline systems, and the spline knots for the mean and variance functions. These knots ensure that the prediction code can regenerate the same spline basis that was used when training the model. Training and prediction Python files (.py) FIles which were used to train models and predict on new, unseen data. Requirements text files (.txt) Package requirements for Python pipeline. README (.txt) Explanation of scripts and their implementation.