ChemLink is a modular orchestration platform for computational chemistry on HPC clusters. Automate docking campaigns and molecular dynamics simulations with a single unified CLI — from GPU-accelerated search to trajectory analysis.
Running molecular docking campaigns and dynamics simulations on HPC infrastructure is complex, fragile, and deeply manual. ChemLink abstracts the orchestration layer so your team can focus on chemistry, not plumbing — reducing experiment setup from 45 minutes to under 5.
Reduces experiment setup time from 45 minutes to under 5. One unified CLI handles job submission, monitoring, and post-processing for both docking and dynamics workflows.
Full pipeline: active site detection with fpocket, grid preparation with AutoGrid4, and conformational search with AutoDock-GPU targeting NVIDIA architectures sm_89, sm_90, and sm_120.
Structured MD simulation from topology preparation to trajectory analysis, supporting six biological system types with CUDA and MPI acceleration via GROMACS 2025.
Both pipelines operate in single-node or distributed mode using SLURM Job Arrays. Scales docking campaigns across multiple GPU nodes with automatic work partitioning.
Individual ligand or trajectory failures are isolated, logged with a full traceback, and skipped without stopping the campaign. A complete success/failure breakdown is reported at the end of every run.
Every run is logged and traceable. Prometheus and Grafana monitoring track cluster health in real time, and Conda-pinned environments guarantee reproducibility across nodes and users.
ChemLink is built in five decoupled layers — CLI, pipelines, steps, adapters, and HPC infrastructure — so each scientific tool can be updated or swapped without touching the orchestration logic.
chemlink docking / chemlink dynamic / chemlink doctor) operating in single-node or SLURM Job Array distributed modeChemLink runs on Linux HPC nodes with NVIDIA GPUs. The installer handles all scientific tool compilation and Conda environment creation.
--full install)The benchmarks reported below were obtained on the following configuration. Any node that meets or exceeds these specs should replicate the published performance.
| Component | Manager | worker1 | worker2 |
|---|---|---|---|
| CPU | Intel Core Ultra 9 285 (24 cores / 24 threads) | Intel Core i9-10900 est. (10 cores / 20 threads) | Intel Core i7-12700 (12 cores / 20 threads) |
| RAM | 32 GB | 64 GB | 32 GB |
| GPU | RTX 5060 Ti 16 GB (Blackwell, sm_120) | RTX 3080 10 GB (Ampere, sm_86) | RTX 3060 LHR 12 GB (Ampere, sm_86) |
| Storage | SK Hynix NVMe SSD 1 TB | Samsung NVMe SSD 1 TB | Samsung NVMe SSD 1 TB |
| Network | Intel GbE + Wi-Fi 7 | Intel I219-LM GbE (1 Gbps) | Intel I219-LM GbE (1 Gbps) |
| OS | Ubuntu 24.04 LTS | Ubuntu 24.04 LTS | Ubuntu 24.04 LTS |
| CUDA | 12.x | 12.x | 12.x |
bash# Installs ChemLink in /opt/chemlink with Conda envs bio + mgl_legacy
curl -fsSL https://raw.githubusercontent.com/PipeJF9/chemlink/main/install.sh | bash
bash# Compiles fpocket, AutoGrid4, AutoDock4, AutoDock-GPU, GROMACS 2025.4
# Estimated: 45–90 min — requires CUDA Toolkit, OpenMPI, cmake
curl -fsSL https://raw.githubusercontent.com/PipeJF9/chemlink/main/install.sh | bash -s -- --full
| Option | Description |
|---|---|
| --full | Compile all scientific tools + GROMACS |
| --with-gromacs | Compile GROMACS only |
| --dir PATH | Install directory (default: /opt/chemlink) |
| --version TAG | Git branch/tag (default: main) |
| --skip-conda | Skip Conda environment creation |
bashchemlink doctor # check environment, GPU, and dependencies
chemlink docking --help # docking pipeline options
chemlink dynamic --help # dynamics pipeline options
ChemLink pipelines are configured via CLI flags at runtime. The chemlink doctor command verifies that the environment is correctly set up before running a pipeline.
| Environment | Contents | Used by |
|---|---|---|
| bio | Python 3.10, ACPYPE, AmberTools, OpenBabel, RDKit, pdbfixer, biopython | Docking & dynamics preparation |
| mgl_legacy | Python 2, MGLTools, pythonsh | AutoDock4 file preparation |
| Variable | Description |
|---|---|
| GMXRC | Path to GROMACS environment script (auto-set by installer) |
| CUDA_VISIBLE_DEVICES | Override visible GPUs for docking runs |
| SLURM_ARRAY_TASK_ID | Set automatically by SLURM for distributed array jobs |
The lab uses two independent shared storage systems with distinct purposes:
| System | Where it runs | Mount point | Purpose |
|---|---|---|---|
| NFS v4 | Linux kernel on manager node | /nfs/chemlink | ChemLink computation — code, Conda envs, inputs, intermediates, and results. Mounted on every compute node; required for all pipeline runs. |
| SMB/CIFS (Samba) | OpenMediaVault NAS node | per-user share | Personal and team storage — accessible from Windows, macOS, and Linux. Used for backups, raw data, and researcher files. Not involved in ChemLink pipelines. |
| Command | Description |
|---|---|
| chemlink docking | Run the molecular docking pipeline (fpocket → AutoGrid4 → AutoDock-GPU) |
| chemlink dynamic | Run the molecular dynamics pipeline (GROMACS 2025.4) |
| Command | Description |
|---|---|
| chemlink doctor | Check environment, GPU availability, and all dependencies |
bash# Single-node docking run
chemlink docking \
--receptor inputs/protein.pdb \
--ligands inputs/ligands/ \
--out results/docking/
# Distributed: SLURM Job Array across 4 GPU nodes
chemlink docking \
--receptor /nfs/chemlink/protein.pdb \
--ligands /nfs/chemlink/ligands/ \
--nodes 4 \
--slurm
bash# Protein-ligand MD simulation with GROMACS
chemlink dynamic \
--system protein-ligand \
--input inputs/complex.pdb \
--out results/md/ \
--gpu --mpi-ranks 8
| Flag | Default | Description |
|---|---|---|
| --receptor PATH | required | Receptor PDB file |
| --ligands PATH | required | Ligand file or directory (.pdbqt / .sdf) |
| --out PATH | ./docking_out | Output directory |
| --nodes N | 1 | SLURM node count for distributed mode |
| --slurm | false | Submit via SLURM Job Array |
| --dry-run | false | Validate inputs without running |
| Flag | Default | Description |
|---|---|---|
| --system TYPE | required | protein | protein-ligand | membrane | protein-membrane | ligand | custom |
| --input PATH | required | Input structure file (.pdb / .gro) |
| --out PATH | ./md_out | Output directory |
| --gpu | false | Enable CUDA GPU acceleration |
| --mpi-ranks N | 1 | MPI rank count for parallel run |
| --slurm | false | Submit via SLURM |
ChemLink is designed for a SLURM-managed cluster with NFS shared storage. All nodes must be reachable via passwordless SSH from the head node. The cluster uses two independent storage systems: NFS v4 served by the Linux kernel on the manager node (exports /nfs/chemlink, mounted on all compute nodes — this is what ChemLink uses), and a dedicated OpenMediaVault NAS node that serves SMB/CIFS/Samba for personal and team file storage accessible from any OS. The two are fully independent and serve different purposes.
bash# Verify cluster connectivity
chemlink doctor --check cluster
# Test NFS mount on all nodes
chemlink doctor --check nfs
Pass --slurm and --nodes N to any pipeline to submit a SLURM Job Array. ChemLink automatically partitions the ligand set or MD replicas across array tasks.
bash# Docking campaign: 1000 ligands across 10 GPU nodes
chemlink docking \
--receptor /nfs/chemlink/targets/cdk2.pdbqt \
--ligands /nfs/chemlink/ligands/set_1000/ \
--nodes 10 \
--slurm \
--partition gpu
AutoDock-GPU and GROMACS are compiled for sm_89 (Ada Lovelace), sm_90 (Hopper), and sm_120 (Blackwell) architectures. Run chemlink doctor to confirm your GPU is detected.
The cluster ships with Prometheus + Grafana for real-time node and GPU monitoring. Access the Grafana dashboard on the head node at port 3000.
CUDA_VISIBLE_DEVICES when running multiple jobs on the same nodechemlink doctor before every campaign to catch config issues earlyChemLink is built in the open at the Laboratorio de Química y Biología Computacional of Universidad del Norte. We welcome bug reports, documentation improvements, and new pipeline contributions.
Fork the repository, create a feature/* or fix/* branch, and open a pull request against develop. All contributions go through code review.
Found a bug or have a feature request? Open an issue with your OS, CUDA version, GPU model, and the full error output from chemlink doctor.
Documentation source lives in /docs in the main repo. Corrections, translations, and new guides are all welcome.
ChemLink's modular architecture makes it straightforward to add new simulation backends or analysis stages. See the developer guide in docs/Desarrollo.md.
ChemLink is released under the MIT License. You are free to use, modify, and distribute this software in any project — commercial or otherwise — provided the copyright notice and permission notice appear in all copies.
Install ChemLink and run your first docking campaign or MD simulation in minutes.
