Data for Deep-learning models of the ascending proprioceptive pathway are subject to illusions

Deep-learning models of the ascending proprioceptive pathway are subject to illusions Authors: Adriana Perez Rotondo*, Merkourios Simos*, Florian David, Sebastian Pigeon, Olaf Blanke, & Alexander Mathis *These authors contributed equally to this work. Affiliation: EPFL - École Polytechnique Fédérale de Lausanne, Switzerland 🇨🇭 Date of upload: May 2025 Link to article: https://www.biorxiv.org/content/10.1101/2025.03.15.643457v1 Link to code: https://github.com/amathislab/ProprioceptiveIllusions ------------------------------- Here we provide the data to run the code provided in github. 1. Data required to train and test the models in `data.zip` ~37 GB when uncompressed. This includes: cleaned_smooth: pre-processed data from FLAG3D and PCR dataset, the elbow flexion datasets used to evaluate the models' performance (EF3D) and effect of vibrations (ES3D). These datasets contain the muscle kinematics used, togther with the spindle models to generate the inputs to the models with spindle inputs. osim: the opensim model used to generate the muscle kinematics for all datasets. spindles: inputs and outputs to the 21 trained models in the paper form the ES3D dataset, used to test tendon vibrations on the models 2. Data to reproduce the figures in the paper in `data_for_figs.zip` ~5.1 GB when uncompressed. 3. Weights for the 21 trained models analyzed in the paper in `trained_models.zip` ~13 MB when uncompressed.