STOMP

STOMP was formerly PSEUDO. The old pseudo-* commands still work as aliases: they print a deprecation warning on stderr and will be removed in STOMP 0.3.0. Before installing, remove the deprecated PSEUDO package: pip uninstall PSEUDO, then delete any leftover deprecated src/PSEUDO.egg-info folder in your checkout. Having both installed breaks either one when the other is removed. Scripts that treat any stderr output as failure should switch to the stomp-* commands.

STOMP (STochastic OMit Perturbation maps) is a computational framework for debiasing and experimental uncertainty quantification in protein structural models resolved by molecular replacement.

It provides a three-stage pipeline that runs on HPC clusters via SLURM:

Stage	Command	What it does
Debias	stomp-debias	Generates stochastic omit perturbation (STOMP) maps via Phenix
Quantify	stomp-quantify	Separates true signal from phase bias across the map ensemble
Analyse	stomp-analyse	Scores every model atom against the debiased SNR map (MUSE)

---

Installation

git clone https://github.com/hamlet-khachatryan/STOMP.git
cd STOMP
pip install -e ".[dev]"

STOMP requires the Phenix Software Suite for STOMP map calculation. Load it before submitting any SLURM jobs:

module load phenix

---

Minimal configuration

Create a YAML file (e.g. run.yaml) with just the required fields:

debias:
  run_name: "my_experiment"
  structure_path: "/data/target.pdb"
  reflections_path: "/data/target.mtz"

paths:
  work_dir: "/scratch/my_project"

slurm:
  partition: "cs05r"

All other parameters fall back to built-in defaults (see Configuration Reference).

---

Input modes

STOMP accepts three input modes for the debias stage:

Mode	Required fields	Behaviour
Single structure	structure_path + reflections_path	Always processed as-is
CSV screening	screening_path (.csv)	All rows always processed; must contain PDB/CIF and MTZ columns
SQLite screening	screening_path (.sqlite)	Diamond SoakDB format; supports outcome filtering and structure count capping

SQLite-specific options

sqlite_outcomes — comma-separated substrings matched against the RefinementOutcome column. Accepted values:

Analysis Pending, PANDDA model - minor, In Refinement,
CompChem ready, Deposition ready, Deposited, Analysed & Rejected

max_structures — cap on the number of structures processed from the SQLite file. Set to null (default) to process all matching entries.

Example config for SQLite with filtering:

debias:
  screening_path: "/data/soakdb.sqlite"
  sqlite_outcomes: "CompChem ready, Deposition ready, Deposited"  # comma-separated string
  max_structures: 50

Or via CLI:

stomp-debias generate-params --config run.yaml \
  --sqlite_outcomes "CompChem ready, Deposited" \
  --max_structures 50

---

Basic usage

Three commands take a structure from raw reflections to per-atom density support scores:

stomp-debias generate-params --config run.yaml

jid=$(sbatch --parsable submit_preprocessing.slurm) && \
sbatch --dependency=afterok:$jid submit_omission.slurm

stomp-quantify --input_path /scratch/my_project/my_experiment
stomp-analyse  --input_path /scratch/my_project/my_experiment

Quantification results are stored in <work_dir>/<run_name>/<crystal>/quantify_results/<k_*_cap_*>/:

File	Description
{stem}_mean.ccp4	STOMP$_{\mu}$ map: average over the STOMP ensemble
{stem}_std.ccp4	STOMP$_{\sigma}$ map: voxel-wise standard deviation map of the STOMP ensemble
{stem}_snr.ccp4	STOMP$_{SNR}$ map: voxel-wise signal-to-noise ratio map
{stem}_p_values.ccp4	voxel-wise signal probability map derived from the STOMP$_{SNR}$ map

Analysis results land in <work_dir>/<run_name>/<crystal>/analyse_results/:

File	Description
{stem}_atoms.csv	Per-atom MUSE scores and diagnostic flags
{stem}_residues.csv	Per-residue MUSEm aggregated scores
{stem}_summary.json	Global statistics (OPIA, counts, thresholds)
{stem}_scored.pdb	Structure with MUSE scores in the B-factor column

Load {stem}_scored.pdb in PyMOL and colour by B-factor to visualise density support across the model.

---

Screen Report

When running on a screening directory (multiple crystals), stomp-analyse automatically generates an interactive HTML summary report at <screening_dir>/index.html.

The report provides:

• Summary cards — total crystals, analysis-complete count, mean OPIA across the run
• Sortable table — one row per crystal with colour-coded OPIA and ligand MUSE badges, SNR significance threshold, links to all output files, and a collapsible binding-site residue table
• Coot integration — one pre-built session script per crystal that loads the scored model, refined 2FoFc map, STOMP_μ map, and STOMP_SNR map, centred on the ligand

以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗