ihmtools 0.0.1a2__tar.gz

ihmtools-0.0.1a2/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Arthur Zalevsky
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

ihmtools-0.0.1a2/MANIFEST.in

@@ -0,0 +1,7 @@
+include README.md
+recursive-include examples *.py *.md
+include examples/data/coordinates.cif examples/data/crosslinks.csv
+include examples/data/restraints.csv examples/data/9A9W.png
+# output, local scratch, and the collection example (11 MB) stay out of the sdist
+exclude examples/data/assembled.cif
+prune examples/data/G_1000003

ihmtools-0.0.1a2/PKG-INFO

@@ -0,0 +1,198 @@
+Metadata-Version: 2.4
+Name: ihmtools
+Version: 0.0.1a2
+Summary: Command-line tools for the PDB-IHM validation and deposition systems
+Author-email: Arthur Zalevsky <aozalevsky@gmail.com>
+License-Expression: MIT
+Project-URL: Homepage, https://github.com/salilab/ihmtools
+Project-URL: Repository, https://github.com/salilab/ihmtools
+Project-URL: PDB-IHM, https://pdb-ihm.org
+Keywords: pdb-ihm,deriva,ermrest,hatrac,integrative modeling
+Classifier: Development Status :: 3 - Alpha
+Classifier: Environment :: Console
+Classifier: Intended Audience :: Science/Research
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Scientific/Engineering :: Bio-Informatics
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: requests
+Provides-Extra: test
+Requires-Dist: pytest>=7; extra == "test"
+Dynamic: license-file
+
+# ihmtools
+
+Command-line tools for the [PDB-IHM](https://pdb-ihm.org) validation and
+deposition systems. They talk to DERIVA's two REST APIs directly — ERMrest for
+records, Hatrac for files — so the only dependency is `requests`.
+
+```bash
+pip install ihmtools
+ihmv login          # once; Globus, via the browser
+```
+
+While only the TestPyPI pre-release exists, the second index is not optional:
+
+```bash
+pip install --index-url https://test.pypi.org/simple/ \
+            --extra-index-url https://pypi.org/simple/ ihmtools
+```
+
+TestPyPI carries its own stale copy of `requests` (2.5.4.1, from 2015), so
+without `--extra-index-url` pip installs that instead of the real one and every
+command dies with `module 'collections' has no attribute 'MutableMapping'`.
+
+Both commands default to the **dev** server; `--mode production` switches.
+
+The examples below live in the repository, so clone it to run them:
+
+```bash
+git clone https://github.com/salilab/ihmtools.git
+cd ihmtools
+pip install -e .
+```
+
+## `ihmv` — validation catalog
+
+```
+ihmv upload model.cif                submit a structure for validation
+ihmv run model.cif                   upload and block until it finishes
+ihmv get_status                      list entries, newest first
+ihmv get_status 2ZJ                  one word plus an exit code
+ihmv set_status 2QJ --to Reprocess   ask the pipeline to run it again
+ihmv download 2Y0 2XT                fetch validation reports
+ihmv delete 2Y0                      remove a record and its reports
+```
+
+## `ihmdep` — deposition system
+
+```
+ihmdep upload model.cif --image model.png      deposit an entry
+ihmdep run model.cif                           deposit and block
+ihmdep get_status                              list entries, newest first
+ihmdep set_status 9-DXAM --to SUBMIT           DRAFT / DEPO / SUBMIT only
+ihmdep download 9-DXAM                         fetch generated reports
+ihmdep delete 9-DXAM                           pre-submit entries only
+```
+
+## Preparing an entry from raw files
+
+`examples/` builds a depositable IHM mmCIF out of what an experimenter
+actually has, using PDB-IHM entry **9A9W** — "USP7 bound to a nucleosome/p53
+complex": histones, two DNA strands, p53, USP7 and four zincs, with DSSO
+crosslinking MS and a 3DEM map.
+
+```bash
+pip install gemmi ihm             # the example needs these two; the CLIs do not
+cd examples                       # from the repository root
+
+python assemble.py                # writes data/assembled.cif
+ihmdep upload data/assembled.cif --image data/9A9W.png
+```
+
+`assemble.py` reads three files and derives everything else with python-ihm —
+entities with the right alphabet for protein, DNA and the zinc ligand, one
+asym unit per chain copy, an atomic representation over the residues actually
+observed, the datasets, both restraints, the protocol, and the model:
+
+| file | what it is |
+|---|---|
+| `data/coordinates.cif` | the model, with no IHM metadata at all |
+| `data/crosslinks.csv` | `id,protein1,residue1,protein2,residue2,linker` |
+| `data/restraints.csv` | `crosslink_id,chain1,chain2` |
+| `data/9A9W.png` | the entry image, for the deposit |
+
+For your own system these come from your pipeline. 9A9W's are checked in, so
+the example runs on a fresh clone; they were recovered from the released entry
+by stripping every `_ihm*` category from `pdb-ihm.org/cif/9A9W.cif`, taking the
+two crosslink tables from `_ihm_cross_link_list` and
+`_ihm_cross_link_restraint`, and fetching `pdb-ihm.org/images/9a9w.png`
+(lowercase id).
+
+The crosslinks are two files because mmCIF keeps them apart and so does the
+science. `crosslinks.csv` is what the experiment measured — protein and
+residue, with no idea which copy. `restraints.csv` is what the modelling
+actually restrained: the chain pair, for the subset used. For 9A9W those are
+90 and 51. Of the 40 measurements left unrestrained, 25 have a residue that
+isn't in the coordinates — it fell in a disordered gap, so there is no atom to
+measure to — and the copy assignment can't be recovered from a measurement at
+all: "H2B residue 24 to H2B residue 28" doesn't say *which* H2B, and there are
+two of each histone and four p53.
+
+The result matches the original where it should: identical atoms and
+sequences, both crosslink categories row for row, and it validates against
+`mmcif_ihm.dic` + `mmcif_pdbx_v50.dic`.
+
+## Scripting
+
+`get_status` exits `0` done, `1` error, `2` pending, `3` unknown RID, so a
+submit-and-wait loop is just:
+
+```bash
+RID=$(ihmv upload model.cif)
+ihmv get_status --wait "$RID" && ihmv download "$RID" -o reports/
+```
+
+`--wait` is a flag; `--interval SECS` changes the 30-second poll. They are
+separate because a RID can be all digits, and an option that took an optional
+value would read `--wait 300` as an interval rather than as RID 300.
+
+Depositing several entries works the same way. `upload` prints nothing but
+the RID on stdout, so the loop's output is the RID list, and every later
+command reads it back with `-`.
+
+`examples/data/G_1000003/` holds three entries from one PDB-IHM collection —
+9A40, 9A6P and 9A7U, from "Modelling protein complexes with crosslinking mass
+spectrometry and deep learning" — with their coordinates and images:
+
+```bash
+cd examples/data/G_1000003        # from the repository root
+
+for id in 9A40 9A6P 9A7U; do
+    ihmdep upload "$id.cif" --image "$id.png"
+done > rids.txt
+
+ihmdep get_status --wait - < rids.txt &&
+ihmdep set_status --to SUBMIT --yes - < rids.txt &&
+ihmdep get_status --wait - < rids.txt &&
+ihmdep download --mmcif -o generated/ - < rids.txt
+```
+
+A failed upload prints no RID, so it drops out of the batch rather than
+stopping it, and re-running the loop picks up the existing RIDs instead of
+depositing twice. `get_status` exits non-zero if any entry errored, which
+keeps a broken batch from being submitted. After SUBMIT the generated mmCIF
+comes first; the validation PDFs arrive later, hence `--mmcif`.
+
+Listings are aligned on a terminal and **tab-separated when piped**, with a
+`#`-prefixed header. Several columns contain spaces (`RECORD READY`, `Error:
+processing uploaded mmCIF file`), so split on tabs rather than whitespace:
+
+```bash
+ihmdep get_status | awk -F'\t' '!/^#/ && $5 ~ /^Error/ {print $1}'
+```
+
+RIDs come from arguments, from `--rid` (repeatable), or from stdin via `-`.
+
+## Notes
+
+The two modules are deliberately self-contained — each can be copied out and
+run on its own — which means they duplicate their auth and HTTP layers. A fix
+to one must be applied to both.
+
+Uploads follow the catalog's own `tag:isrd.isi.edu,2017:asset` annotation for
+where files go and which extensions are accepted, which is what the web UI
+obeys. Don't substitute the `bulk-upload` annotation that `deriva-upload-cli`
+reads: on dev it points at a different Hatrac namespace.
+
+## Tests
+
+From the repository root:
+
+```bash
+pip install -e '.[test]'
+pytest
+```
+
+Offline only — no network or credentials needed.

ihmtools-0.0.1a2/README.md

@@ -0,0 +1,175 @@
+# ihmtools
+
+Command-line tools for the [PDB-IHM](https://pdb-ihm.org) validation and
+deposition systems. They talk to DERIVA's two REST APIs directly — ERMrest for
+records, Hatrac for files — so the only dependency is `requests`.
+
+```bash
+pip install ihmtools
+ihmv login          # once; Globus, via the browser
+```
+
+While only the TestPyPI pre-release exists, the second index is not optional:
+
+```bash
+pip install --index-url https://test.pypi.org/simple/ \
+            --extra-index-url https://pypi.org/simple/ ihmtools
+```
+
+TestPyPI carries its own stale copy of `requests` (2.5.4.1, from 2015), so
+without `--extra-index-url` pip installs that instead of the real one and every
+command dies with `module 'collections' has no attribute 'MutableMapping'`.
+
+Both commands default to the **dev** server; `--mode production` switches.
+
+The examples below live in the repository, so clone it to run them:
+
+```bash
+git clone https://github.com/salilab/ihmtools.git
+cd ihmtools
+pip install -e .
+```
+
+## `ihmv` — validation catalog
+
+```
+ihmv upload model.cif                submit a structure for validation
+ihmv run model.cif                   upload and block until it finishes
+ihmv get_status                      list entries, newest first
+ihmv get_status 2ZJ                  one word plus an exit code
+ihmv set_status 2QJ --to Reprocess   ask the pipeline to run it again
+ihmv download 2Y0 2XT                fetch validation reports
+ihmv delete 2Y0                      remove a record and its reports
+```
+
+## `ihmdep` — deposition system
+
+```
+ihmdep upload model.cif --image model.png      deposit an entry
+ihmdep run model.cif                           deposit and block
+ihmdep get_status                              list entries, newest first
+ihmdep set_status 9-DXAM --to SUBMIT           DRAFT / DEPO / SUBMIT only
+ihmdep download 9-DXAM                         fetch generated reports
+ihmdep delete 9-DXAM                           pre-submit entries only
+```
+
+## Preparing an entry from raw files
+
+`examples/` builds a depositable IHM mmCIF out of what an experimenter
+actually has, using PDB-IHM entry **9A9W** — "USP7 bound to a nucleosome/p53
+complex": histones, two DNA strands, p53, USP7 and four zincs, with DSSO
+crosslinking MS and a 3DEM map.
+
+```bash
+pip install gemmi ihm             # the example needs these two; the CLIs do not
+cd examples                       # from the repository root
+
+python assemble.py                # writes data/assembled.cif
+ihmdep upload data/assembled.cif --image data/9A9W.png
+```
+
+`assemble.py` reads three files and derives everything else with python-ihm —
+entities with the right alphabet for protein, DNA and the zinc ligand, one
+asym unit per chain copy, an atomic representation over the residues actually
+observed, the datasets, both restraints, the protocol, and the model:
+
+| file | what it is |
+|---|---|
+| `data/coordinates.cif` | the model, with no IHM metadata at all |
+| `data/crosslinks.csv` | `id,protein1,residue1,protein2,residue2,linker` |
+| `data/restraints.csv` | `crosslink_id,chain1,chain2` |
+| `data/9A9W.png` | the entry image, for the deposit |
+
+For your own system these come from your pipeline. 9A9W's are checked in, so
+the example runs on a fresh clone; they were recovered from the released entry
+by stripping every `_ihm*` category from `pdb-ihm.org/cif/9A9W.cif`, taking the
+two crosslink tables from `_ihm_cross_link_list` and
+`_ihm_cross_link_restraint`, and fetching `pdb-ihm.org/images/9a9w.png`
+(lowercase id).
+
+The crosslinks are two files because mmCIF keeps them apart and so does the
+science. `crosslinks.csv` is what the experiment measured — protein and
+residue, with no idea which copy. `restraints.csv` is what the modelling
+actually restrained: the chain pair, for the subset used. For 9A9W those are
+90 and 51. Of the 40 measurements left unrestrained, 25 have a residue that
+isn't in the coordinates — it fell in a disordered gap, so there is no atom to
+measure to — and the copy assignment can't be recovered from a measurement at
+all: "H2B residue 24 to H2B residue 28" doesn't say *which* H2B, and there are
+two of each histone and four p53.
+
+The result matches the original where it should: identical atoms and
+sequences, both crosslink categories row for row, and it validates against
+`mmcif_ihm.dic` + `mmcif_pdbx_v50.dic`.
+
+## Scripting
+
+`get_status` exits `0` done, `1` error, `2` pending, `3` unknown RID, so a
+submit-and-wait loop is just:
+
+```bash
+RID=$(ihmv upload model.cif)
+ihmv get_status --wait "$RID" && ihmv download "$RID" -o reports/
+```
+
+`--wait` is a flag; `--interval SECS` changes the 30-second poll. They are
+separate because a RID can be all digits, and an option that took an optional
+value would read `--wait 300` as an interval rather than as RID 300.
+
+Depositing several entries works the same way. `upload` prints nothing but
+the RID on stdout, so the loop's output is the RID list, and every later
+command reads it back with `-`.
+
+`examples/data/G_1000003/` holds three entries from one PDB-IHM collection —
+9A40, 9A6P and 9A7U, from "Modelling protein complexes with crosslinking mass
+spectrometry and deep learning" — with their coordinates and images:
+
+```bash
+cd examples/data/G_1000003        # from the repository root
+
+for id in 9A40 9A6P 9A7U; do
+    ihmdep upload "$id.cif" --image "$id.png"
+done > rids.txt
+
+ihmdep get_status --wait - < rids.txt &&
+ihmdep set_status --to SUBMIT --yes - < rids.txt &&
+ihmdep get_status --wait - < rids.txt &&
+ihmdep download --mmcif -o generated/ - < rids.txt
+```
+
+A failed upload prints no RID, so it drops out of the batch rather than
+stopping it, and re-running the loop picks up the existing RIDs instead of
+depositing twice. `get_status` exits non-zero if any entry errored, which
+keeps a broken batch from being submitted. After SUBMIT the generated mmCIF
+comes first; the validation PDFs arrive later, hence `--mmcif`.
+
+Listings are aligned on a terminal and **tab-separated when piped**, with a
+`#`-prefixed header. Several columns contain spaces (`RECORD READY`, `Error:
+processing uploaded mmCIF file`), so split on tabs rather than whitespace:
+
+```bash
+ihmdep get_status | awk -F'\t' '!/^#/ && $5 ~ /^Error/ {print $1}'
+```
+
+RIDs come from arguments, from `--rid` (repeatable), or from stdin via `-`.
+
+## Notes
+
+The two modules are deliberately self-contained — each can be copied out and
+run on its own — which means they duplicate their auth and HTTP layers. A fix
+to one must be applied to both.
+
+Uploads follow the catalog's own `tag:isrd.isi.edu,2017:asset` annotation for
+where files go and which extensions are accepted, which is what the web UI
+obeys. Don't substitute the `bulk-upload` annotation that `deriva-upload-cli`
+reads: on dev it points at a different Hatrac namespace.
+
+## Tests
+
+From the repository root:
+
+```bash
+pip install -e '.[test]'
+pytest
+```
+
+Offline only — no network or credentials needed.

ihmtools-0.0.1a2/examples/README.md

@@ -0,0 +1,43 @@
+# Building a 9A9W entry
+
+`assemble.py` turns raw files into a depositable IHM mmCIF, as a worked
+example of preparing an entry. **See the main README for the workflow and what
+the input files are** — this covers only what is worth knowing if you adapt
+the code.
+
+From the repository root:
+
+```bash
+pip install gemmi ihm
+cd examples && python assemble.py
+```
+
+One entry, deliberately — not a general converter. It reads
+`data/coordinates.cif`, `data/crosslinks.csv` and `data/restraints.csv`, and
+writes `data/assembled.cif`. gemmi reads the coordinates; python-ihm describes
+everything else.
+
+The inputs are checked in, so this runs on a fresh clone. Only the output and
+local scratch are ignored.
+
+## Things that bite
+
+- **`_entity.pdbx_description` is the only naming a coordinate file carries.**
+  Without it the crosslink list's protein names have nothing to match against.
+  gemmi's `make_mmcif_document()` drops it, so anything that rewrites the
+  coordinates has to put it back.
+- **The alphabets are keyed differently.** `DNAAlphabet` uses `DA`,
+  `LPeptideAlphabet` the one-letter code. Look each component up by name
+  first, then by letter.
+- **Non-polymers have no `label_seq`.** The zincs vanish from the model if you
+  skip atoms without one, and the write then fails with *"Assemblies reference
+  asym IDs that don't have coordinates"*. A single-component entity's `seq_id`
+  is always 1.
+- **Ask the entity whether it is polymeric** before giving an asym unit a
+  residue range — older gemmi assigns `label_seq` to non-polymers too, and
+  `ihm` rejects a range on a ligand.
+- **Chains have gaps.** Chain A holds 96 of its 139 residues, so the
+  representation describes what was modelled, not the full sequence.
+
+Needs current gemmi and python-ihm (tested with gemmi 0.7.5 / ihm 2.11);
+gemmi 0.5.8 groups entities differently and the build fails.

ihmtools-0.0.1a2/examples/assemble.py

@@ -0,0 +1,212 @@
+#!/usr/bin/env python3
+"""Rebuild PDB-IHM entry 9A9W from raw coordinates and a crosslink list.
+
+9A9W is "USP7 bound to a nucleosome/p53 complex": histones, two DNA strands,
+p53, USP7 and four zincs, with DSSO crosslinking MS and a 3DEM map. It reads
+data/coordinates.cif, data/crosslinks.csv and data/restraints.csv, and writes
+data/assembled.cif.
+
+gemmi reads the coordinates; python-ihm describes everything else. This is a
+worked example for one entry, not a general converter -- the metadata below is
+9A9W's, and the code assumes what 9A9W contains.
+"""
+
+import csv
+import os
+
+import gemmi
+
+import ihm
+import ihm.dataset
+import ihm.dumper
+import ihm.location
+import ihm.model
+import ihm.protocol
+import ihm.reader
+import ihm.representation
+import ihm.restraint
+
+DATA = os.path.join(os.path.dirname(os.path.abspath(__file__)), "data")
+COORDINATES = os.path.join(DATA, "coordinates.cif")
+CROSSLINKS = os.path.join(DATA, "crosslinks.csv")
+RESTRAINTS = os.path.join(DATA, "restraints.csv")
+OUTPUT = os.path.join(DATA, "assembled.cif")
+
+# --------------------------------------------------------------------------
+# Read the coordinates. gemmi gives the sequences, the chain grouping and the
+# entity descriptions; those descriptions are what the crosslink list names
+# its proteins by.
+# --------------------------------------------------------------------------
+
+structure = gemmi.read_structure(COORDINATES)
+structure.setup_entities()
+structure.assign_label_seq_id()
+
+block = gemmi.cif.read(COORDINATES).sole_block()
+category = block.get_mmcif_category("_entity")
+descriptions = dict(zip(category["id"], category["pdbx_description"]))
+
+# --------------------------------------------------------------------------
+# One Entity per unique sequence, one AsymUnit per chain copy.
+#
+# The alphabets are keyed differently -- DNAAlphabet by "DA", LPeptideAlphabet
+# by the one-letter code -- so look each component up by name first. The four
+# zincs are non-polymers; each needs its own single-component entity.
+# --------------------------------------------------------------------------
+
+entities = {}                                  # gemmi entity name -> Entity
+asyms = {}                                     # subchain id -> AsymUnit
+
+for gemmi_entity in structure.entities:
+    description = descriptions[gemmi_entity.name]
+
+    if gemmi_entity.entity_type == gemmi.EntityType.Polymer:
+        alphabet = (ihm.DNAAlphabet
+                    if gemmi_entity.polymer_type == gemmi.PolymerType.Dna
+                    else ihm.LPeptideAlphabet)
+        comps = alphabet()._comps
+        sequence = [
+            comps.get(name)
+            or comps[gemmi.find_tabulated_residue(name).one_letter_code.upper()]
+            for name in gemmi_entity.full_sequence]
+        entity = ihm.Entity(sequence, alphabet=alphabet, description=description)
+    else:
+        name = next(residue.name for chain in structure[0] for residue in chain
+                    if residue.subchain in gemmi_entity.subchains)
+        entity = ihm.Entity([ihm.NonPolymerChemComp(name, name=description)],
+                            description=description)
+
+    entities[gemmi_entity.name] = entity
+    for subchain in gemmi_entity.subchains:
+        asyms[subchain] = ihm.AsymUnit(entity, details=description, id=subchain)
+
+by_description = {e.description: e for e in entities.values()}
+by_chain = {a.id: a for a in asyms.values()}
+
+system = ihm.System(title="USP7 bound to a nucleosome/p53 complex")
+system.authors.extend(["Chakraborty, D.", "Kempf, G.", "Kater, L.",
+                       "Cavadini, S.", "Thoma, N.H."])
+system.entities.extend(entities.values())
+system.asym_units.extend(asyms.values())
+
+assembly = ihm.Assembly(list(asyms.values()), name="Modeled assembly")
+
+# --------------------------------------------------------------------------
+# Representation: atomic, over the residues actually present. Chains have gaps
+# -- chain A holds 96 of its 139 residues -- so describe what was modelled
+# rather than the whole sequence.
+# --------------------------------------------------------------------------
+
+observed = {}
+for chain in structure[0]:
+    for residue in chain:
+        if residue.label_seq is not None:
+            observed.setdefault(residue.subchain, []).append(residue.label_seq)
+
+# A residue range only means anything for a polymer; the zincs take the whole
+# asym unit. (Ask the entity rather than checking for seq ids -- older gemmi
+# assigns label_seq to non-polymers too.)
+representation = ihm.representation.Representation([
+    ihm.representation.AtomicSegment(
+        asym(min(observed[sid]), max(observed[sid]))
+        if asym.entity.is_polymeric() else asym,
+        rigid=False)
+    for sid, asym in asyms.items()])
+
+# --------------------------------------------------------------------------
+# The data 9A9W was built from.
+# --------------------------------------------------------------------------
+
+crosslink_data = ihm.dataset.CXMSDataset(ihm.location.PRIDELocation("PXD054141"))
+em_data = ihm.dataset.EMDensityDataset(ihm.location.EMDBLocation("EMD-53517"))
+system.orphan_datasets.extend([
+    crosslink_data, em_data,
+    ihm.dataset.DeNovoModelDataset(
+        ihm.location.AlphaFoldDBLocation("AF-Q93009-F1-v4")),
+    ihm.dataset.PDBDataset(ihm.location.PDBLocation("9R04")),
+])
+
+# --------------------------------------------------------------------------
+# Crosslinks come in two layers, as they do in mmCIF.
+#
+# crosslinks.csv is what the experiment measured: protein and residue, with no
+# idea which copy. restraints.csv is what the modelling actually restrained --
+# the chain pair, for the subset of measurements used. 9A9W measures 90 and
+# restrains 51 of them; the other 40 stay in the list and restrain nothing.
+# --------------------------------------------------------------------------
+
+crosslink_restraint = ihm.restraint.CrossLinkRestraint(
+    dataset=crosslink_data, linker=ihm.ChemDescriptor("DSSO"))
+
+experimental = {}                              # measurement id -> link
+for row in csv.DictReader(open(CROSSLINKS)):
+    link = ihm.restraint.ExperimentalCrossLink(
+        by_description[row["protein1"]].residue(int(row["residue1"])),
+        by_description[row["protein2"]].residue(int(row["residue2"])))
+    experimental[row["id"]] = link
+    crosslink_restraint.experimental_cross_links.append([link])
+
+distance = ihm.restraint.UpperBoundDistanceRestraint(30.0)
+for row in csv.DictReader(open(RESTRAINTS)):
+    crosslink_restraint.cross_links.append(ihm.restraint.ResidueCrossLink(
+        experimental_cross_link=experimental[row["crosslink_id"]],
+        asym1=by_chain[row["chain1"]], asym2=by_chain[row["chain2"]],
+        distance=distance))
+
+em_restraint = ihm.restraint.EM3DRestraint(
+    dataset=em_data, assembly=assembly, fitting_method="Flexible fitting")
+system.restraints.extend([crosslink_restraint, em_restraint])
+
+# --------------------------------------------------------------------------
+# How the model was built.
+# --------------------------------------------------------------------------
+
+protocol = ihm.protocol.Protocol(name="Modeling")
+for name, method in [("Structure prediction", "AlphaFold2"),
+                     ("Rigid body fitting", "Rigid body fitting"),
+                     ("Model editing", "Manual editing"),
+                     ("Flexible fitting", "Molecular dynamics flexible fitting"),
+                     ("Refinement", "Maximum-likelihood refinement")]:
+    protocol.steps.append(ihm.protocol.Step(
+        assembly=assembly, dataset_group=None, method=method, name=name,
+        num_models_begin=None, num_models_end=None))
+system.orphan_protocols.append(protocol)
+
+
+class Model(ihm.model.Model):
+    """Streams atoms from gemmi rather than holding a second copy of them."""
+
+    def get_atoms(self):
+        for chain in structure[0]:
+            for residue in chain:
+                # Non-polymers have no label_seq, but a single-component
+                # entity's seq_id is always 1.
+                seq_id = residue.label_seq or 1
+                for atom in residue:
+                    yield ihm.model.Atom(
+                        asym_unit=asyms[residue.subchain], seq_id=seq_id,
+                        atom_id=atom.name, type_symbol=atom.element.name,
+                        x=atom.pos.x, y=atom.pos.y, z=atom.pos.z,
+                        het=residue.het_flag == "H",
+                        biso=atom.b_iso, occupancy=atom.occ)
+
+
+model = Model(assembly=assembly, protocol=protocol,
+              representation=representation, name="Best scoring model")
+em_restraint.fits[model] = ihm.restraint.EM3DRestraintFit()
+
+system.state_groups.append(ihm.model.StateGroup(
+    [ihm.model.State([ihm.model.ModelGroup([model], name="All models")])]))
+
+with open(OUTPUT, "w") as fh:
+    ihm.dumper.write(fh, [system])
+print("wrote %s" % OUTPUT)
+
+# Read it back -- the cheapest check that what we wrote is well formed.
+with open(OUTPUT) as fh:
+    check, = ihm.reader.read(fh)
+crosslinks = check.restraints[0]
+print("%d entities, %d restraints, %d measurements, %d restrained"
+      % (len(check.entities), len(check.restraints),
+         sum(len(g) for g in crosslinks.experimental_cross_links),
+         len(crosslinks.cross_links)))