This repository contains the codebase for an efficient workflow designed for processing electrophysiological data. The workflow is specifically tailored for analyzing in silico generated data from MEArec.
-
Data Processing: The workflow facilitates seamless processing of electrophysiological data originating from MEArec. This includes preprocessing steps to prepare data for further analysis.
-
AI-based Spike Detection: The core of this project features a powerful Transformer model customized for spike detection in neurophysiological time series data. This allows advanced analysis and identification of relevant events in the data.
-
Evaluation and visualization: Evaluation and visualization functions for further investigation of the development and analyzing process.
-
Import: Handling different file formats in order to apply the trained model.
The project was developed under Python 3.9 and Linux Ubuntu 22.04 lts. For machine learning tasks a GPU is recommended.
To use the workflow, follow these steps:
- Clone the repository:
git clone https://github.com/tivenide/SpikeSense - Install the required dependencies:
pip install -r requirements.txt
Please install MEArec separately and follow their instructions on: https://mearec.readthedocs.io/en/latest/installation.html
Execute data processing with the following command:
python cli/spike_detection.py path/to/MEArec_recording_input_data.h5 path/to/spike_trains_output_dataOnly for demonstration purposes to better understand the workflow and check the installation. Please adjust your MEArec data according to your requirement.
- Use
data_generation_MEArec/MEArec_data_generation_quick.pyto generate some demo files. - Put the recordings for training into
quick/MEArec_training_data_recordings.
Execute development (quick) with the following command:
python cli/quick_cli.py developExecute data processing with the following command:
python cli/quick_cli.py process path/to/MEArec_recording_input_data.h5 path/to/spike_trains_output_data- choosing license
Copyright 2023 tivenide. All rights reserved.