The Virtual Lab is an AI-human collaboration for science research. In the Virtual Lab, a human researcher works with a team of large language model (LLM) agents to perform scientific research. Interaction between the human researcher and the LLM agents occurs via a series of team meetings, where all the LLM agents discuss a scientific agenda posed by the human researcher, and individual meetings, where the human researcher interacts with a single LLM agent to solve a particular scientific task.
Please see our paper The Virtual Lab: AI Agents Design New SARS-CoV-2 Nanobodies with Experimental Validation for more details on the Virtual Lab and an application to nanobody design for SARS-CoV-2.
As a real-world demonstration, we applied the Virtual Lab to design nanobodies for one of the latest variants of SARS-CoV-2 (see nanobody_design). The Virtual Lab built a computational pipeline consisting of ESM, AlphaFold-Multimer, and Rosetta and used it to design 92 nanobodies that were experimentally validated.
Please see the notebook nanobody_design/run_nanobody_design.ipynb for an example of how to use the Virtual Lab to create agents and run team and individual meetings.
The Virtual Lab can be installed using pip or by cloning the repo and installing the required packages. Installation should only take a couple of minutes.
Optionally, first create a conda environment.
conda create -y -n virtual_lab python=3.12
conda activate virtual_lab
The Virtual Lab can be installed via pip.
pip install virtual-lab
To install a local version of the Virtual Lab, clone the repo and then install the package.
git clone https://github.com/zou-group/virtual_lab.git
cd virtual_lab
pip install -e .
The Virtual Lab currently uses GPT-4o from OpenAI. Save your OpenAI API key as the environment variable OPENAI_API_KEY
. For example, add export OPENAI_API_KEY=<your_key>
to your .bashrc
or .bash_profile
.