Skip to content
Martini-Force-Field-InitiativePublic

About

Python package for computing partition coefficients of small molecules using Martini 3 coarse-grained molecular dynamics

Resources

Stars

3 stars

Watchers

1 watching

Forks

Repository files navigation

Taster

taster is a Python package for computing partition coefficients of Martini 3 small molecules using free energy calculations.

Given a molecule's CG structure and ITP file, taster automates the full workflow:

  • Solvation box preparation across multiple solvents
  • Parallel FEP/TI simulations via GROMACS
  • Free energy estimation using TI or MBAR
  • LogP calculation relative to a reference solvent (typically water)
  • Convergence and overlap diagnostics (forward/backward convergence, MBAR overlap matrix, TI dhdl plots) for each leg

Requirements

  • Python >= 3.11
  • GROMACS (2024.3 or greater needed if topologies use angles with restricted bending potentials)
  • numpy >= 1.20
  • pandas >= 1.0
  • matplotlib >= 3.5
  • MDAnalysis >= 2.0
  • alchemlyb >= 2.0
  • tqdm >= 4.60
  • jax >= 0.7.2

Development dependencies are managed in uv dependency groups. The test group includes pytest and alchemtest; lint, typecheck, and docs provide the corresponding development tools.


Installation

Install uv first.

For Users

Install taster with the dependencies needed to run and test it:

git clone https://github.com/Martini-Force-Field-Initiative/taster.git
cd taster
uv sync --group test

Add as well --extra cuda for NVIDIA GPU support or --extra rocm for AMD GPU support

This creates a project environment in .venv using the versions pinned in uv.lock. GROMACS must also be installed separately.

For Developers

Install taster with all linting, type-checking, testing, and documentation tools:

uv sync --all-groups

Add as well --extra cuda for NVIDIA GPU support or --extra rocm for AMD GPU support


Quick Start

Python API

import taster

# Run the full workflow
df = taster.run_partition_workflow(
    itp="MOL.itp",
    structure="MOL.gro",
    solvents=["water", "octanol-water_74-26", "hexadecane", "chloroform"],
    reference="water",
    reps=3,
    ncores=36,
)

print(df)

Or run each step individually:

import taster

# Step 1 — prepare simulation boxes
resname = taster.prepare_partition_setup(
    itp="MOL.itp",
    structure="MOL.gro",
    solvents=["water", "octanol-water_74-26", "hexadecane"],
    reps=3,
)

# Step 2 — run TI simulations
taster.run_partitions(
    resname, solvents=["water", "octanol-water_74-26", "hexadecane"], reps=3, ncores=36
)

# Step 3 — analyse
df = taster.process_partition(
    resname, solvents=["octanol-water_74-26", "hexadecane"], water="water", reps=3
)

CLI

taster --itp MOL.itp --structure MOL.gro \
       --solvents water octanol-water_74-26 hexadecane chloroform \
       --reps 3 --ncores 36 --estimator MBAR

See all options:

taster --help

Available Solvents

import taster

print(taster.available_solvents())

Output

Results are saved to ./Partitions/<resname>/:

  • results.txt — human-readable report of LogP and ΔG values per replicate and average
  • partition_avg.npy — numpy array of averaged results (dG, dG_err, logP, logP_err) per solvent

process_partition and run_partition_workflow also return a pandas.DataFrame:

rep  solvent               dG      dG_err   logP    logP_err
1    octanol-water_74-26  -12.3    0.4      2.15    0.07
1    hexadecane           -15.1    0.5      2.64    0.09
...
avg  octanol-water_74-26  -12.3    0.2      2.16    0.04
avg  hexadecane           -15.1    0.3      2.64    0.05

Convergence and Overlap Diagnostics

By default (diagnostics=True), process_partition also saves convergence/overlap figures for every replicate and solvent leg (including the reference solvent) under ./Partitions/<resname>/diagnostics/<rep>/<solvent>/:

  • mbar_convergence.png / ti_convergence.png — forward/backward convergence of the free energy estimate
  • mbar_overlap_matrix.png — MBAR overlap matrix between neighbouring lambda states (MBAR only)
  • ti_dhdl.png — dHdl curve across lambda states (TI only)
  • convergence.log — convergence analysis log, kept out of the terminal

Pass diagnostics=False to skip this and only compute LogP values.


Testing

To run the full test suite from the main folder run:

uv run pytest -v tests/

This runs all tests in tests/; alchemtest is installed through the uv test dependency group.


Notes on HPC Usage

By default ncores is set to os.cpu_count(), which returns the total number of logical CPUs on the machine. On shared HPC nodes this will be the node's full core count.

Always pass ncores explicitly to match the resources available to your job.


Developed by:


License

LGPLv2.1 License. See LICENSE for details.

About

Python package for computing partition coefficients of small molecules using Martini 3 coarse-grained molecular dynamics

Resources

Stars

3 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages