Contrastive-Machine-Learning-reveals-the-structure-of-individual-variation-in-ASD

Code and materials for paper "Contrastive machine learning reveals the structure of neuroanatomical variation within Autism"

• Notebooks/

  • 01-Train-AutoEncoders.ipynb Trains VAE and CVAE autoencoders. For more info see Abid & Zou, 2019
  • 02-Extract-Latent-Features.ipynb Extracts latent features (shared, ASD-specific & VAE) using trained autoencoder models
  • 03-Analysis-RSA.ipynb RSA analyses (Figure 1B/1C)
  • 04-Analysis-Clustering-Results.ipynb Clustering analyses (Figure 1D)
  • 05-Jacobian-Make-Jacobians.ipynb Generates the Jacobian Determinant maps using synthethic "TC-Twins"
  • 06-Jacobian-Jacobian-Analysis.ipynb Calculates LOSO PCA and plots neuroanatomical associations (Figure 2)
  • helper_funcs.py helper functions called by analysis notebooks
  • make_models2.py Code defining VAE and CVAE Tensorflow models

• Data/

  • Generated data and necessary files

• tf_weights/ Trained weights for VAE and CVAE models

  • CVAE_weights/
    • CVAE_weights.z01
    • CVAE_weights.z02
    • ...
  • VAE_weights/
    • VAE_weights.z01
    • VAE_weights.z02
    • ...

N.B. The trained weights are zipped into a multi-part zip file (.z01,.z02,.z03 etc.) and need to be unzipped before use.

N.B. SFARI data has been removed to comply with Researcher Distribution Agreement (eRDA). SFARI Award #614379 to Joshua Hartshorne and Stefano Anzelotti. Data can be obtained by authorized researchers from base.sfari.org