Huntianling

English | 简体中文 | 视频

Huntianling is an integrated multi-agent skills system designed for AI-driven drug discovery (AIDD), covering the full computational discovery workflow from target protein investigation, structural analysis, pocket-based molecular generation, docking, molecular dynamics (MD) simulations, to high-accuracy FEP free-energy calculations. Rather than functioning as isolated tools, the Agents act as clearly specialized “professional AI roles” that can collaborate with one another: they automatically hand off tasks within a unified framework, share intermediate results, and interact with team members through natural language and structured information. This enables a modularized, automated, and collaborative research workflow, advancing drug discovery from “human-driven” to “agent-collaboration-driven.”

Skills

• Bioinformatics & structural investigation: UniProt/PDB fetch, chain & conformation selection, identification of missing residues and ligands.

• Patent research: WO/US/CN patent searching and analysis

• PDB file processing: PDB download, format conversion, chain splitting, hydrogen addition/completion, protonation, removal of waters/ions, etc.

• Protein preparation: conformation cleanup, residue repair, apo/holo selection, energy minimization

• Molecular dynamics (MD) simulation: system setup, short relaxation, production runs, trajectory and energy output/analysis

• Pocket prediction: geometry/energy/deep-learning based scoring

• Molecule generation: SMILES generation based on pocket/constraints/fragments/property conditions

• Ligand preparation: desalting, stereoisomers/protonation/tautomers selection, 3D conformers, SDF output

• Molecular docking: grid docking box, docking, scoring, pose filtering/selection

• FEP: relative/absolute free-energy calculation, FEP workflow management, topology mapping, results summarization

• Synthetic route prediction: retrosynthetic routes analysis

• Communication & collaboration: email drafting (internal medicinal chemistry review / external vendor RFQs)

• Compound registration: structured entry of compound & project metadata into inventory (registration number, batch, properties, source, etc.)

Installation

1. Clone the project

git clone https://github.com/Gewu-Intelligence/Huntianling
cd Huntianling

2. Create a conda environment

conda env create -f environment.yml
conda activate huntianling

3. Install OpenCode

npm install -g opencode-ai
## or
curl -fsSL https://opencode.ai/install | bash

You can also install it using other methods described in the OpenCode docs.

After that, you should be able to launch the OpenCode TUI (terminal user interface) from your shell:

opencode

or start a local web server:

OPENCODE_SERVER_USERNAME=who OPENCODE_SERVER_PASSWORD=secret opencode web --hostname 127.0.0.1 --port 4059

Open a browser and go to [https://localhost:4059]. Log in with the corresponding username and password to access OpenCode’s web service. For LAN access, replace 127.0.0.1 with 0.0.0.0.

For the LLM model, Zhipu AI GLM-4.7 is recommended.

4. Other software (optional)

1. Gromacs

   If you plan to use GROMACS for MD simulations, make sure gmx or gmx_mpi is correctly set in your environment variables (PATH). Please refer to theGROMACS Installation guide

   # vi ~/.bashrc
   # settings of MPI environmnet
   # settings of cuda if it is required 
   source $GMX_PATH/install/bin/GMXRC
   gmx_mpi --version

2. Retrosynthesis

   For retrosynthesis analysis, use the team-developed RXNGraphormer

   conda create -n rxngraphormer python=3.10
   conda activate rxngraphormer
   git clone -b pytorch2 https://github.com/licheng-xu-echo/RXNGraphormer.git
   cd RXNGraphormer/
   
   pip install torch==2.2.1 --index-url https://download.pytorch.org/whl/cu121
   pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.2.0+cu121.html
   pip install rdkit==2024.3.2 ipykernel pandas python-box OpenNMT-py==1.2.0 torchdata==0.7.1 torch_geometric rxnmapper localmapper transformers==4.30.0 numpy==1.26.4 scikit-learn
   pip install .

Example

More examples can be found in example docs

• Download pdb

Download the PDB file of 8S99 (save as .pdb，output to ./pdb)

• Structure preparation

Protein preparation of 8S99 (input ./pdb/8S99.pdb，output to ./pdb)

• MD

Run MD simulation (10000 steps, input ./pdb/protein_A_apo.pdb.pdb；output to ./md)

• Pocket prediction

Predict the pocket of 8S99, (input ./pdb/protein_A_apo.pdb.pdb, output to ./pocket/)