From 5f2676e33a4318f6183fb8b9c02b7e5e9400d551 Mon Sep 17 00:00:00 2001
From: Stefan Bienert <stefan.bienert@unibas.ch>
Date: Tue, 22 Aug 2023 14:32:50 +0200
Subject: [PATCH] Get draft of ModelCIF 'header'
---
convert_to_modelcif.py | 162 ++++++++++++++++++++++++++++++++++-------
pyproject.toml | 8 ++
2 files changed, 145 insertions(+), 25 deletions(-)
diff --git a/convert_to_modelcif.py b/convert_to_modelcif.py
index 398ff2c..bd2a21f 100755
--- a/convert_to_modelcif.py
+++ b/convert_to_modelcif.py
@@ -3,11 +3,15 @@
"""Take the output of the AlphaPulldown pipeline and turn it into a ModelCIF
file with a lot of metadata in place."""
+from typing import Tuple
+import json
import os
import sys
from absl import app, flags, logging
+import modelcif
+import modelcif.dumper
# ToDo: Get options properly, best get the same names as used in existing
# scripts, e.g. could '--monomer_objects_dir' be used as feature
@@ -30,44 +34,144 @@ FLAGS = flags.FLAGS
# exist as expected.
-def alphapulldown_model_to_modelcif() -> None:
+def _store_as_modelcif(
+ data_json: dict,
+ mdl_file: str,
+ out_dir: str,
+ # ost_ent, file_prfx, compress, add_files
+) -> None:
+ """Create the actual ModelCIF file."""
+ system = modelcif.System(
+ title=data_json["_struct.title"],
+ id=data_json["data_"].upper(),
+ model_details=data_json["_struct.pdbx_model_details"],
+ )
+
+ # write modelcif.System to file
+ # NOTE: this will dump PAE on path provided in add_scores
+ # -> hence we cheat by changing path and back while being exception-safe...
+ oldpwd = os.getcwd()
+ os.chdir(out_dir)
+ try:
+ with open(
+ f"{os.path.splitext(os.path.basename(mdl_file))[0]}.cif",
+ "w",
+ encoding="ascii",
+ ) as mmcif_fh:
+ modelcif.dumper.write(mmcif_fh, [system])
+ # _package_associated_files(system.repositories[0])
+ # if compress:
+ # _compress_cif_file(mdl_fle)
+ finally:
+ os.chdir(oldpwd)
+
+
+def _get_model_details(cmplx_name: str, data_json: dict) -> str:
+ """Get the model description."""
+ ap_versions = []
+ for mnmr in data_json["__meta__"]:
+ if data_json["__meta__"][mnmr]["ap_version"] not in ap_versions:
+ ap_versions.append(data_json["__meta__"][mnmr]["ap_version"])
+
+ # ToDo: fetch AF2 version/ have it in metadata JSON
+ return (
+ f"Model generated for {' and '.join(cmplx_name)}, produced "
+ + "using AlphaFold-Multimer (<AF2 VERSION>) as implemented by "
+ + f"AlphaPulldown ({', '.join(ap_versions)})."
+ )
+
+
+def _get_feature_metadata(
+ modelcif_json: dict, cmplx_name: str, prj_dir: str
+) -> list:
+ """Read metadata from a feature JSON file."""
+ cmplx_name = cmplx_name.split("_and_")
+ prj_dir = os.path.join(prj_dir, "features_monomers")
+ if not os.path.isdir(prj_dir):
+ logging.info(f"No feature directory '{prj_dir}' found.")
+ sys.exit()
+ if "__meta__" not in modelcif_json:
+ modelcif_json["__meta__"] = {}
+ for mnmr in cmplx_name:
+ modelcif_json["__meta__"][mnmr] = {}
+ feature_json = os.path.join(prj_dir, f"{mnmr}_feature_metadata.json")
+ if not os.path.isfile(feature_json):
+ logging.info(f"No feature metadata file '{feature_json}' found.")
+ sys.exit()
+ # ToDo: make sure that its always ASCII
+ with open(feature_json, "r", encoding="ascii") as jfh:
+ jdata = json.load(jfh)
+ modelcif_json["__meta__"][mnmr]["ap_version"] = jdata["version"]
+
+ return cmplx_name
+
+
+def _get_data_block_id_and_struct_and_entry_categories(
+ cif_json: dict, cmplx_name: str
+) -> None:
+ """Get 'data_' block ID and data for categories '_struct' and '_entry'."""
+ cif_json["data_"] = "_".join(cmplx_name)
+ cif_json["_struct.title"] = f"Prediction for {' and '.join(cmplx_name)}"
+ cif_json["_struct.pdbx_model_details"] = _get_model_details(
+ cmplx_name, cif_json
+ )
+
+
+def alphapulldown_model_to_modelcif(
+ cmplx_name: str,
+ mdl_file: str,
+ out_dir: str,
+ prj_dir: str,
+) -> None:
"""Convert an AlphaPulldown model into a ModelCIF formatted mmCIF file.
Metadata for the ModelCIF categories will be fetched from AlphaPulldown
output as far as possible. This expects modelling projects to exists in
AlphaPulldown's output directory structure."""
-
-
-def _get_model_list(ap_dir, model_selected) -> list:
+ # ToDo: ENABLE logging.info(f"Processing '{mdl_file}'...")
+ modelcif_json = {}
+ # fetch metadata
+ cmplx_name = _get_feature_metadata(modelcif_json, cmplx_name, prj_dir)
+ # fetch/ assemble more data about the modelling experiment
+ _get_data_block_id_and_struct_and_entry_categories(
+ modelcif_json, cmplx_name
+ )
+ _store_as_modelcif(modelcif_json, mdl_file, out_dir)
+ # ToDo: ENABLE logging.info(f"... done with '{mdl_file}'")
+
+
+def _get_model_list(ap_dir: str, model_selected: str) -> Tuple[str, str, list]:
"""Get the list of models to be converted.
If `model_selected` is none, all models will be marked for conversion."""
# ToDo: Question - use 'ranked_*.pdb' or
# 'unrelaxed_model_*_multimer_v3_pred_0.pdb' models?
+ mdl_path = os.path.join(ap_dir, "models")
+ cmplx = os.listdir(mdl_path)
+ # For now, exactly 1 complex is expected in the 'models' subdirectory. If
+ # there are more, the 'model_selected' mechanism needs to be further tuned
+ # to get to the right model.
+ assert len(cmplx) == 1
+ cmplx = cmplx[0]
+ mdl_path = os.path.join(mdl_path, cmplx)
+ models = []
if model_selected is not None:
- mdl_paths = os.path.join(ap_dir, "models")
- cmplx = os.listdir(mdl_paths)
- # For now, exactly 1 complex is expected in the 'models' subdirectory.
- # If there are more, the 'model_selected' mechanism needs to be further
- # tuned to get to the right model.
- assert len(cmplx) == 1
- cmplx = cmplx[0]
- mdl_paths = os.path.join(
- mdl_paths, cmplx, f"ranked_{model_selected}.pdb"
- )
- mdl_paths = [mdl_paths]
+ models.append(os.path.join(mdl_path, f"ranked_{model_selected}.pdb"))
+ else:
+ for mdl in os.listdir(mdl_path):
+ if mdl.startswith("ranked_"):
+ models.append(os.path.join(mdl_path, mdl))
- for mdl in mdl_paths:
- if os.path.isfile(mdl):
+ # check that files actually exist
+ for mdl in models:
+ if not os.path.isfile(mdl):
logging.info(
f"Model file '{mdl}' does not exist or is not a regular file."
)
sys.exit()
- # check that files actually exist
-
- return []
+ return cmplx, mdl_path, models
def main(argv):
@@ -92,10 +196,14 @@ def main(argv):
# pylint: enable=pointless-string-statement
del argv # Unused.
- # make list of selected models
- model_conversions = _get_model_list(FLAGS.ap_output, FLAGS.model_selected)
- # assemble selected models into ModelCIF files + associated data archives
- alphapulldown_model_to_modelcif()
+ # get list of selected models and assemble ModelCIF files + associated data
+ complex_name, model_dir, model_list = _get_model_list(
+ FLAGS.ap_output, FLAGS.model_selected
+ )
+ for mdl in model_list:
+ alphapulldown_model_to_modelcif(
+ complex_name, mdl, model_dir, FLAGS.ap_output
+ )
if __name__ == "__main__":
@@ -106,5 +214,9 @@ if __name__ == "__main__":
# but we did that already in the past. Idea is to have all models
# available for... reproducibility and whatnot, but show the selected
# (representative) of the modelling experiment/ study more prominently.
+# ToDo: Things to look at: '_struct.title', '_struct.pdbx_model_details',
+# 'data_', '_entry', maybe have a user-defined JSON document with things
+# like that, including author names?
+# ToDo: where to store which model was chosen? Should be in Tara's models.
-# LocalWords: ToDo AlphaPulldown PAEs dir
+# LocalWords: ToDo AlphaPulldown PAEs dir struct
diff --git a/pyproject.toml b/pyproject.toml
index c80443d..f9bbc98 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -9,8 +9,16 @@ dependencies = [
[tool.black]
line-length = 80
+[tool.pylint.MAIN]
+load-plugins = "pylint.extensions.bad_builtin"
+
[tool.pylint.REPORTS]
reports = "no"
[tool.pylint.FORMAT]
max-line-length = 81
+
+[tool.pylint.deprecated_builtins]
+# We want to use proper logging, so we can control *ALL* output bei the Abseil
+# logger, hence: deprecate 'print'
+bad-functions = ["map", "filter", "print"]
\ No newline at end of file
--
GitLab