What is quantumdrug-agent?

QuantumDrug Agent is an open source research platform that tries to automate the early stages of drug discovery by chaining together large language models, a team of specialized AI agents, chemistry tools, and a quantum inspired search technique. Rather than only generating candidate molecule shapes, the project aims to walk through the fuller research process, starting from picking a disease target and ending with a written scientific report explaining the reasoning behind the final candidates. The system works as a pipeline of specialized agents. A research agent looks up validated drug targets from the ChEMBL public database. A hypothesis agent then proposes treatment ideas based on that biological information. A generator agent creates candidate molecules using RDKit, a chemistry toolkit. A quantum agent uses a technique called Matrix Product States, borrowed from physics, to compress the huge space of possible molecules and rank promising candidates faster than checking everything one by one. Further agents estimate how well a candidate might bind to its target, evaluate drug like properties such as absorption and toxicity, rank the candidates, and finally explain the reasoning and write up a report. An interactive assistant lets you ask questions about the results afterward. The project is built in Python with a Streamlit web interface, using RDKit for chemistry, the ChEMBL API for target data, Plotly for interactive charts, and a locally run Llama 3.1 model through Ollama for the language reasoning parts, alongside a tensor network library called Quimb for the quantum inspired ranking step. Installing it means cloning the repository, installing the Python requirements, and running the Streamlit app. The README lists an example workflow for heart failure as a sample disease input, but notes a roadmap of unfinished items including real molecular docking, protein structure support, and GPU acceleration, meaning several described capabilities are still planned rather than complete. It is released under the MIT license.