Beyond propulsion: muscle proprioception enables hydrodynamic sensing in fish body

Project: Beyond propulsion: muscle proprioception enables hydrodynamic sensing in fish body Journal: Proceedings of the Royal Society BAuthors: Afridi et al.DOI: doi.org/10.1098/rspb.2025.0474. Repository Contents 1. Code File/ Contains all analysis scripts and raw data needed to reproduce the results. a) Matlab code/ Dataset_actual_joint_angles.(csv|mat) – Ground-truth joint angle data extracted from video-based kinematics. Dataset_BPNN_pred_joint_angles.(csv|mat) – Predicted joint angles from the BPNN model. Dataset_CNN_pred_joint_angles.(csv|mat) – Predicted joint angles from the CNN model. Dataset_LSTM_pred_joint_angles.(csv|mat) – Predicted joint angles from the LSTM model. FishPoseDLCtoJointSpace.m – Script to convert DeepLabCut (DLC) outputs into joint space angles. FishPosefromMLModelsPrediction.m – Script to reconstruct full-body pose from ML-predicted joint angles. detect_EMGmaxima_dynamic_window.m – Detect EMG onset peaks using adaptive windowing. instantaneous_phase_shift.m – Compute phase alignment between predicted and measured joint angles. rmse.m – Utility to calculate Root Mean Square Error (RMSE). TimeOrderAnalysisOfEMGandKinematicsSFcndtnWholeBodyNewDevice.m – Main script for computing Δt (delay) between EMG onset and body displacement (used in proprioceptive sensing analysis). RMSE_0.274.csv – Example RMSE results at 0.274 m/s swimming speed. Test DLC filtered data.csv, Test EMG raw data.csv – Example datasets used for validation. b) Python code/ 1 BPNN code.py – Neural network model (Backpropagation NN). 2 LSTM code.py – Neural network model (Long Short-Term Memory). 3 CNN code.py – Neural network model (Convolutional NN). CustomFunctions.py – Helper functions for preprocessing and model training. FilterAndFeatureExtraction.py – Preprocessing script for EMG signals: filtering and feature extraction. c) EMG and Kinematic Data/ Contains synchronized raw EMG and kinematic recordings for all flow conditions. LaminarFlowDATA/ – EMG and kinematics for control (steady laminar flow). 5cmD/ – EMG and kinematics for fish swimming behind 5 cm D-section cylinder (Kármán vortices). 7cmD/ – EMG and kinematics for fish swimming behind 7 cm D-section cylinder. File naming convention: EMG SBJ#_Condition.csv – EMG data. Kin SBJ#_Condition.csv – Corresponding kinematic data. EMG Stream_YYYY_MM_DD_HHMMSS.csv / Kin Stream_YYYY_MM_DD_HHMMSS.csv – Continuous stream data recordings. How Files Relate to Manuscript Results Figure 2 (ML validation): Generated using 1 BPNN code.py, 2 LSTM code.py, 3 CNN code.py on data from LaminarFlowDATA/. Output datasets: Dataset_BPNN_pred_joint_angles, Dataset_CNN_pred_joint_angles, Dataset_LSTM_pred_joint_angles. Validation metrics (RMSE, phase alignment) computed with rmse.m and instantaneous_phase_shift.m. Figure 3 (Temporal Δt analysis): Uses TimeOrderAnalysisOfEMGandKinematicsSFcndtnWholeBodyNewDevice.m on data from LaminarFlowDATA/, 5cmD/, and 7cmD/. Input: EMG and kinematic .csv files. Output: Distributions of Δt delays used in Figure 3c, f, i. Supplementary Figures S18–S19 (CFD comparison): Not reproduced here (done via IBAMR simulations by collaborators), but synchronized EMG and kinematic datasets are provided for linking experimental recordings to CFD conditions. Software Requirements MATLAB (R2020a or later) – for temporal analysis, RMSE, Δt calculation, pose reconstruction. Python (3.8 or later) – for neural network training and validation. Dependencies: TensorFlow/Keras, NumPy, pandas, matplotlib (versions compatible with Python 3.8). Instructions for Reproducing Key Results ML-based validation (Figure 2): Run FilterAndFeatureExtraction.py to preprocess EMG signals. Train models using 1 BPNN code.py, 2 LSTM code.py, and 3 CNN code.py. Evaluate predictions with rmse.m and instantaneous_phase_shift.m. Body pose reconstruction: Run FishPoseDLCtoJointSpace.m to generate joint angles from kinematic data. Apply FishPosefromMLModelsPrediction.m to reconstruct body pose from ML-predicted angles. Temporal analysis of proprioception (Figure 3): Use detect_EMGmaxima_dynamic_window.m to identify EMG onset peaks. Run TimeOrderAnalysisOfEMGandKinematicsSFcndtnWholeBodyNewDevice.m with EMG and kinematic data to compute Δt distributions. Citation If you use these data or scripts, please cite:Rahdar Hussain Afridi et al., “Beyond propulsion: muscle proprioception enables hydrodynamic sensing in fish body,” Proceedings of the Royal Society B, 2025, doi: https://doi.org/10.1098/rspb.2025.0474.