This repository contains the source code of the ICNet model and an example Jupyter notebook ICNet_example.ipynb that can be used to simulate neural activity in the inferior colliculus (IC) in response to a sound stimulus. A Google Colab version of the notebook can also be found here and can be executed without a local Python installation. The supporting work can be found on bioRxiv and can be cited as follows:
Drakopoulos, F., Sabesan, S., Xia, Y., Fragner, A. & Lesica, N. A. (2024). Modeling neural coding in the auditory midbrain with high resolution and accuracy. bioRxiv 2024.06.17.599294; doi: 10.1101/2024.06.17.599294
The DNN folder contains the deep-neural-network (DNN) architecture, weights and configuration parameters of the pre-trained ICNet model. The src folder contains all the supplementary Python functions to execute ICNet. This repository also contains an example sound file scribe_male_talker.wav from the UCL SCRIBE dataset, a pre-compiled Python version of the provided Jupyter notebook ICNet_example.py, this README.MD document and a license file.
ICNet was developed to describe the shared sound encoding across a set of 9 normal-hearing gerbils. The output_to_simulate variable in the example notebook can be adjusted to simulate neural activity of individual animals (e.g. output_to_simulate='animal_random' for a random animal), to generate neural responses using units from all 9 animals (e.g. output_to_simulate='units_1000' for 1000 neural units selected from all animals with logarithmic spacing of CFs along the IC), or to derive the response of the shared bottleneck layer for use as a front-end (output_to_simulate='bottleneck'). The characteristic frequencies (CFs) of all neural units are included in the DNN folder to simulate neural responses that reflect the tonotopic organization of the IC.
To run the example Jupyter notebook and the ICNet model, it is necessary to install the Python packages included in requirements.txt. We recommend installing conda and executing the following commands:
conda create --name ICNet python=3.11
conda activate ICNet
pip install -r requirements.txt
The first two commands are used to create (and activate) a new conda environment named ICNet with Python v3.11. If using only Python (without conda), running the last command is sufficient to install all necessary packages.
If you use this code, please cite the corresponding paper or this repository:
Drakopoulos, F., Fragner, A. & Lesica, N. A. (2024). ICNet: A DNN model of central auditory processing (v1.0). Zenodo. https://zenodo.org/doi/10.5281/zenodo.11842064.
For questions, please reach out to one of the corresponding authors:
- Fotios Drakopoulos: [email protected]
- Nicholas A Lesica: [email protected]
This work was supported by EPSRC EP/W004275/1 and MRC MR/W019787/1.