Spike Agent

Overview

SpikeAgent is a multimodal large language model (LLM)-based AI agent that automates and standardizes the spike sorting pipeline, a critical process in neural signal decoding. Traditional workflows are fragmented, labor-intensive, and require expert curation, limiting scalability and reproducibility. SpikeAgent overcomes these challenges by leveraging LLMs for autonomous decision-making, reasoning, and interaction with intermediate results.

Features

• End-to-End Automation – Handles preprocessing, spike detection, feature extraction, clustering, and validation.
• Multimodal Interaction – Supports text, voice, and image-based inputs.
• Autonomous Reasoning – Uses LLMs and coding functions to optimize sorting without human intervention.
• Interpretable Results – Generates detailed reports explaining sorting decisions.
• High Efficiency – Reduces curation time from 40+ minutes to just a few minutes per 100 spikes.
• Benchmark-Validated – Matches or surpasses expert-level curation across various spike-sorting algorithms and electrode types.

Getting Started

1. Clone the repository

   git clone https://github.com/LiuLab-Bioelectronics-Harvard/SpikeAgent.git
   cd SpikeAgent

2. Install the dependencies

   We use conda to manage the dependencies and currently only support the Linux system.

   Linux Users

   # Create the environment from the file
   conda env create -f environment.linux.yml
   
   # Activate the environment
   conda activate spike_agent

3. Set up your environment variables

   • Create a .env file in the root directory
   • Add your API key in this file. The API keys needed for this agent are:

   # If you want to use OpenAI models, you need to add the following:
   OPENAI_API_KEY=your-api-key-here
   OPENAI_API_BASE=your-api-base-here
   WHISPER_API_KEY=your-whisper-api-key-here
   
   # If you want to use Claude, you need to add the following:
   ANTHROPIC_API_KEY=your-anthropic-api-key-here
   
   # If you want to use Gemini, you need to add the following:
   GOOGLE_API_KEY=your-google-api-key-here
   

4. (Optional) Modify the default parameters in src/Agent/default.py

   • Set the raw_data_path to the path of your raw data.
   • Set the processed_path to the path of your processed data.
   • Set the folder_name to the data directory name under your raw data path.

   You may also directly share parameters with the agent and our agent will set them for you.

5. Run the app

   streamlit run src/unified_app.py

The app will open in your default web browser at the local host.

Usage

1. src: this directory contains the source code of the spike sorting agent.
2. data: this directory can be used to hold the data for the spike sorting. It also contains an example dataset in raw_data/0330. You can unzip the data files and testing with our agent.
3. src/tmp/plots: this directory contains the plots generated by the agent.
4. conversation_histories_{model}: these directories contains the conversation history of the agent classfied by the model used. You may load and save the conversation history with the agent.

Citation

If you find SpikeAgent useful for your work, please cite our paper:

Zuwan Lin#, Arnau Marin-Llobet#, Jongmin Baek, Yichun He, Jaeyong Lee, Wenbo Wang, Xinhe Zhang, Ariel J. Lee, Ningyue Liang, Jin Du, Jie Ding, Na Li, Jia Liu* Preprint at bioRxiv (2025): https://doi.org/10.1101/2025.02.11.637754