FastHydroMap

FastHydroMap predicts residue-level hydrophobicity (Fdewet) and water-structure maps (PC1, PC2, PC3) from protein structures. It runs on a CPU without explicit-water simulations. Lower Fdewet indicates a more hydrophobic local surface.

Quick start

In a fresh Python environment (Python 3.11–3.14):

python -m pip install fasthydromap
fasthydromap install-torch
fasthydromap predict protein.pdb -o results/protein

This writes one CSV with all four scores, four PDBs with the scores in their B-factor columns, and an NPZ file of residue embeddings for downstream analysis.

For missing heavy atoms, add --prepare and inspect the saved prepared structure.

ChimeraX

In ChimeraX, open Tools → More Tools…, find FastHydroMap in the Toolshed, and click Install. Use bundle 0.2.1 or newer and restart ChimeraX after upgrading an existing bundle. Run fasthydromap install once to set up its Python dependencies, then predict and color a structure:

fasthydromap #1
fasthydromap #1 quantity pc1

See the ChimeraX tutorial for installation, coloring output PDBs, and interpreting the maps.

Trajectories

fasthydromap predict-trajectory protein.pdb trajectory.dcd -o results/trajectory

DCD and XTC trajectory files are supported. This writes trajectory.npz with all four scores per frame and residue, plus residue IDs and frame/time information. Add --csv for a wide _total.csv, or --embeddings to include per-frame embeddings in the NPZ. Make sure you have a complete topology with matching atom order, and make molecules whole before prediction. See how to load the results.

Inputs and scope

Use standard protein heavy atoms. Ordinary HIS defaults to neutral HIE; HID/HIE/HIP names or --histidine-states states.csv specify an assignment. FastHydroMap does not estimate pKa. See input details.

Training used restrained, single-chain protein simulations. Buried residues, termini, assemblies, and unsupported chemistry need caution.

Documentation and citation

• Installation
• Python API, inputs, embeddings, and model
• Training and data
• Protein list and dataset splits
• HydroMap labels, structures and training inputs

Lobo, S.; Najafi, S.; Shea, J.-E.; Shell, M. S. Context-Aware Hydrophobicity Modeling: HydroMap and FastHydroMap. bioRxiv (2026). doi:10.64898/2026.06.07.730647.

Shell Lab · Shea Group