The latest local version of CVSP-AIE, designated as ​​HierVS​​ (Hierarchical Virtual Screening), is now accessible via GitHub. HierVS represents a ​​fully AI-driven hierarchical virtual screening platform​​, specifically engineered to support ​​ultra-large-scale virtual screening workflows​​. HierVS first employs KarmaDock for efficient docking, followed by CarsiDock for accurate docking and RTMscore for re-scoring. HierVS facilitates straightforward local deployment and execution, with dynamically adjustable virtual screening duration based on the setting of precise-docked molecular quantities.

System Requirements

Component Specification
Operating System Ubuntu 22.04.5
Processor Intel(R) Core(TM) i9-14900KF
GPU NVIDIA GeForce RTX 4090
Storage Minimum 24 GB of available hard-disk space
CUDA Version 12.4
NVIDIA GPU Driver Version 550.144.03
Docker Version 28.0.1
Package Management PyPI (Python package management tool)

The software and hardware listed above are recommended, as our testing was conducted in this environment. Other software or hardware configurations may also be capable of running this tool properly. In addition to the System Requirements mentioned above, Ubuntu24.04 and Ubuntu20.04, CUDA Version 12.8 cooperated with NVIDIA GPU Driver Version 570.133.07 and CUDA Version 11.4 cooperated with NVIDIA GPU Driver Version 470.82.01 have been verified and confirmed to work without issues.

Installation

Step 1: Download Docker Image

(Download time depends on your network speed)

Hugging Face:

wget https://huggingface.co/gushukai/HierVS/resolve/main/hier_vs_v9.tar

Step 2: Deploy the docker image

(Takes several seconds to minutes)

docker load -i hier_vs_v9.tar

Step 3: Install HierVS package

(Takes several seconds)

pip install HierVS

Conducting hierarchical Virtual screening

Virtual screening can be performed using HierVS with the following command:

HierVS -p protein_file -l ligand_file -cl compound_library_file -o output_path -n TopN -v IMAGE_VERSION

Where:

  • -p protein_file: Path to the protein structure file (PDB format)
  • -l ligand_file: Path to the reference ligand file (SDF or MOL2 format)
  • -cl compound_library_file: Path to the compound library file (TXT format)
  • -o output_path: Path to the output directory
  • -n TopN: Number of top molecules for CarsiDock and RTMScore
  • -v IMAGE_VERSION: Docker image version (default: hier_vs:v9)
📝 Note: Important Advisory on RDKit Version

In the CarsiDock molecular docking pipeline, a known issue in specific versions of RDKit can, on rare occasions, cause a segmentation fault during the generation of initial molecular conformations. This may lead to unexpected termination of the screening task.

Recommended Solution: If you encounter this error, please switch the Docker image to hier_vs:v9b. This version contains an upgraded RDKit installation that specifically resolves this stability issue. Prior to use, first download and deploy the image using:

wget https://huggingface.co/gushukai/HierVS/resolve/main/hier_vs_v9b.tar
docker load -i hier_vs_v9b.tar

⚠️ Key Change: hier_vs:v9 → hier_vs:v9b

Example:

HierVS -p ./examples/example_protein.pdb -l ./examples/example_ligand.sdf -cl ./examples/example_compound_library.txt -o ./examples/result -n 50 -v hier_vs:v9

Expected results:

  • karmadock_prediction.csv: Contains KarmaDock prediction results
  • carsidock_prediction.csv: Contains RTMScore rescoring results based on CarsiDock-based docking poses
  • *.sdf files: Ligand poses generated by CarsiDock

For more information, run HierVS --help.

Loading structure...