|
| 1 | +""" |
| 2 | +Annotating and Saving Protein Structures |
| 3 | +========================================= |
| 4 | +
|
| 5 | +This example walks through how to add custom annotations to AtomArrays, visualize them, and save them for later use. |
| 6 | +
|
| 7 | +**Prerequisites**: Familiarity with :doc:`load_and_visualize_structures` for basic structure loading and exploration. |
| 8 | +
|
| 9 | +.. figure:: /_static/examples/annotate_and_save_structures_01.png |
| 10 | + :alt: Heme pocket visualization |
| 11 | + :width: 400px |
| 12 | +
|
| 13 | + Visualization of heme-binding pocket atoms (within 6Å of heme ligand) in myoglobin. |
| 14 | +""" |
| 15 | + |
| 16 | +######################################################################## |
| 17 | +# Setup and Structure Loading |
| 18 | +# ---------------------------- |
| 19 | +# |
| 20 | +# Let's start by loading a protein structure that we'll annotate. We'll use the same myoglobin structure from the loading example: |
| 21 | + |
| 22 | +import os |
| 23 | +import tempfile |
| 24 | + |
| 25 | +import biotite.structure as struc |
| 26 | +import numpy as np |
| 27 | + |
| 28 | +from atomworks.io import parse |
| 29 | +from atomworks.io.utils.io_utils import to_cif_file |
| 30 | +from atomworks.io.utils.testing import get_pdb_path_or_buffer |
| 31 | +from atomworks.io.utils.visualize import view |
| 32 | + |
| 33 | +# sphinx_gallery_thumbnail_path = '_static/examples/annotate_and_save_structures_01.png' |
| 34 | + |
| 35 | +# Load myoglobin structure with heme |
| 36 | +example_pdb_id = "101m" # Myoglobin with heme |
| 37 | +pdb_path = get_pdb_path_or_buffer(example_pdb_id) |
| 38 | + |
| 39 | +# Parse the structure (no need to add missing atoms, since we would just remove them in the following step) |
| 40 | +atom_array = parse(pdb_path, add_missing_atoms=False, fix_formal_charges=False)["assemblies"]["1"][0] |
| 41 | + |
| 42 | +print(f"Loaded structure with {len(atom_array)} atoms") |
| 43 | +print(f"Chains: {np.unique(atom_array.chain_id)}") |
| 44 | + |
| 45 | +# Clean up coordinates (remove any NaN values, if present) |
| 46 | +# (NaN coordinates will break our later step when we create a CellList with Biotite) |
| 47 | +valid_coords_mask = ~np.isnan(atom_array.coord).any(axis=1) |
| 48 | +atom_array = atom_array[valid_coords_mask] |
| 49 | +print(f"After removing NaN coordinates: {len(atom_array)} atoms") |
| 50 | + |
| 51 | +######################################################################## |
| 52 | +# Adding Custom Annotations |
| 53 | +# -------------------------- |
| 54 | +# |
| 55 | +# Now let's add custom annotations to mark different types of atoms. We'll use pocket identification as an example to demonstrate how to create meaningful structural annotations for many ML and general bioinformatics applications. |
| 56 | +# |
| 57 | +# Step 1: Identify Structural Features (Pocket Identification) |
| 58 | +# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 59 | +# |
| 60 | +# Let's efficiently identify the heme-binding pocket using spatial distance cutoffs with Biotite's ``CellList`` class: |
| 61 | + |
| 62 | +# Find atoms within 6 Angstroms of the heme using a spatial cell list |
| 63 | +cell_list = struc.CellList(atom_array.coord, cell_size=6.0) |
| 64 | +heme_coords = atom_array.coord[atom_array.res_name == "HEM"] |
| 65 | + |
| 66 | +print(f"Found {len(heme_coords)} heme atoms") |
| 67 | + |
| 68 | +# Get all atoms within 6Å of any heme atom |
| 69 | +pocket_mask = cell_list.get_atoms(heme_coords, 6.0, as_mask=True) |
| 70 | +pocket_mask = np.any(pocket_mask, axis=0) # Combine results for all heme atoms |
| 71 | + |
| 72 | +print(f"Found {np.sum(pocket_mask)} atoms within 6Å of heme") |
| 73 | + |
| 74 | +# %% |
| 75 | + |
| 76 | +# Visualize the pocket region (always a helpful sanity-check, and trivial with AtomWorks) |
| 77 | +print("\nVisualizing pocket region (all atoms within 6Å of heme):") |
| 78 | +view(atom_array[pocket_mask]) |
| 79 | + |
| 80 | +######################################################################## |
| 81 | +# .. figure:: /_static/examples/annotate_and_save_structures_01.png |
| 82 | +# :alt: Heme pocket visualization |
| 83 | + |
| 84 | +######################################################################## |
| 85 | +# Step 2: Create Annotations from Identified Features |
| 86 | +# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 87 | +# |
| 88 | +# Now we'll convert our pocket identification into an explicit ``AtomArray`` annotation and visualize it: |
| 89 | + |
| 90 | +# Boolean annotation for pocket residues (excluding heme itself) |
| 91 | +is_pocket = pocket_mask & (atom_array.res_name != "HEM") |
| 92 | +atom_array.set_annotation("is_hem_pocket", is_pocket.astype(bool)) |
| 93 | + |
| 94 | +# Boolean annotation for heme atoms |
| 95 | +is_heme = atom_array.res_name == "HEM" |
| 96 | +atom_array.set_annotation("is_heme", is_heme.astype(bool)) |
| 97 | + |
| 98 | +print(f" - Pocket atoms: {np.sum(atom_array.is_hem_pocket)}") |
| 99 | +print(f" - Heme atoms: {np.sum(atom_array.is_heme)}") |
| 100 | + |
| 101 | +# %% |
| 102 | + |
| 103 | +# Visualize just the pocket residues |
| 104 | +print("\nVisualizing annotated pocket residues:") |
| 105 | +view(atom_array[atom_array.is_hem_pocket]) |
| 106 | + |
| 107 | +######################################################################## |
| 108 | +# .. figure:: /_static/examples/annotate_and_save_structures_02.png |
| 109 | +# :alt: Annotated pocket residues visualization |
| 110 | + |
| 111 | +######################################################################## |
| 112 | +# Saving Annotated Structures |
| 113 | +# ---------------------------- |
| 114 | +# |
| 115 | +# Now let's save our annotated structure. In many use cases we may want to save our modified ``AtomArray`` to disk and later load again, preserving our original annotations. |
| 116 | +# |
| 117 | +# AtomWorks provides two methods to do so: |
| 118 | +# |
| 119 | +# .. list-table:: |
| 120 | +# :header-rows: 0 |
| 121 | +# |
| 122 | +# * - Saving to CIF, adding extra annotations directly into the file |
| 123 | +# * - Standard Python object pickling (which may be sensitive to versions, libraries, etc.) |
| 124 | +# |
| 125 | +# Saving to CIF Files |
| 126 | +# ~~~~~~~~~~~~~~~~~~~ |
| 127 | +# |
| 128 | +# CIF files are the standard for structural data and allow us to store arbitrary annotations and categories. |
| 129 | + |
| 130 | +# Create temporary directory for our files |
| 131 | +temp_dir = tempfile.mkdtemp() |
| 132 | +print(f"Working in temporary directory: {temp_dir}") |
| 133 | + |
| 134 | +# Save to CIF file with custom annotations specified |
| 135 | +cif_path = os.path.join(temp_dir, "annotated_structure.cif") |
| 136 | +custom_fields = ["is_hem_pocket", "is_heme"] |
| 137 | + |
| 138 | +saved_cif_path = to_cif_file( |
| 139 | + atom_array, |
| 140 | + cif_path, |
| 141 | + extra_fields=custom_fields, |
| 142 | +) |
| 143 | + |
| 144 | +print(f"Saved CIF file to: {saved_cif_path}") |
| 145 | +print(f"File size: {os.path.getsize(saved_cif_path) / 1024:.1f} KB") |
| 146 | + |
| 147 | +######################################################################## |
| 148 | +# Note on Biological Assemblies and CIF Saving |
| 149 | +# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 150 | +# |
| 151 | +# In some cases, you may find that ``to_cif_file`` reports an error when the structure represents a biological assembly containing multiple copies of the asymmetric unit. The reason for this error is that ``AtomWorks`` builds the biological assembly and explicitly represents every atom; we can't then reverse that process since we may be left with ambiguous bond annotations (e.g., no way to distinguish between multiple copies of "Chain A"). The best solution is to either (a) set the ``chain_id`` to the ``chain_iid`` (which resolves the ambiguity) or (b) simply save the object using a pickle. |
| 152 | +# |
| 153 | +# More rigorous solutions exist; a helpful place for contributions! |
| 154 | +# |
| 155 | +# Alternative Storage Options |
| 156 | +# ~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 157 | +# |
| 158 | +# For Python-specific workflows, you can also save structures as pickle files to preserve exact data types, though CIF files are recommended for interoperability and long-term storage. |
| 159 | + |
| 160 | +######################################################################## |
| 161 | +# Loading Annotated Structures |
| 162 | +# ----------------------------- |
| 163 | +# |
| 164 | +# When we load pickled ``AtomArray``'s, we should restore our original object out-of-the-box with all annotations preserved. |
| 165 | +# |
| 166 | +# When loading from CIF, however, we may need to grapple with data type issues, since within CIF files all fields are considered strings. |
| 167 | +# |
| 168 | +# In the future, we would like to automatically detect annotation data types during loading (and/or allow specification of data types) - we would love contributions and a PR! |
| 169 | +# |
| 170 | +# Loading from CIF Files |
| 171 | +# ~~~~~~~~~~~~~~~~~~~~~~ |
| 172 | + |
| 173 | +from atomworks.io.utils.io_utils import load_any |
| 174 | + |
| 175 | +# Load from CIF file |
| 176 | +loaded_from_cif = load_any(saved_cif_path, extra_fields="all")[0] |
| 177 | + |
| 178 | +print("Loaded from CIF file:") |
| 179 | +print(f" Atoms: {len(loaded_from_cif)}") |
| 180 | +print(" Custom annotations:") |
| 181 | +for annotation in loaded_from_cif.get_annotation_categories(): |
| 182 | + if annotation in custom_fields: |
| 183 | + dtype = getattr(loaded_from_cif, annotation).dtype |
| 184 | + print(f" ✓ {annotation} ({dtype})") |
| 185 | + |
| 186 | +######################################################################## |
| 187 | +# Handling Data Type Conversions |
| 188 | +# ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 189 | +# |
| 190 | +# As we can see above, when boolean annotations are saved to CIF files, they become string representations ("True"/"False"). Here's how to convert them back (we welcome contributions to automate this process and/or allow explicit specification): |
| 191 | + |
| 192 | + |
| 193 | +# Convert string booleans back to actual boolean type |
| 194 | +def fix_boolean_annotation(atom_array: struc.AtomArray, annotation_name: str) -> struc.AtomArray: |
| 195 | + """Convert string boolean annotations back to bool type.""" |
| 196 | + string_values = getattr(atom_array, annotation_name) |
| 197 | + boolean_values = string_values == "True" |
| 198 | + atom_array.del_annotation(annotation_name) |
| 199 | + atom_array.set_annotation(annotation_name, boolean_values) |
| 200 | + return atom_array |
| 201 | + |
| 202 | + |
| 203 | +# Fix boolean annotations |
| 204 | +loaded_from_cif = fix_boolean_annotation(loaded_from_cif, "is_hem_pocket") |
| 205 | +loaded_from_cif = fix_boolean_annotation(loaded_from_cif, "is_heme") |
| 206 | + |
| 207 | +print("\nAfter conversion:") |
| 208 | +print(f" is_hem_pocket: {loaded_from_cif.is_hem_pocket.dtype}, {np.sum(loaded_from_cif.is_hem_pocket)} True values") |
| 209 | +print(f" is_heme: {loaded_from_cif.is_heme.dtype}, {np.sum(loaded_from_cif.is_heme)} True values") |
| 210 | +print(f" Sample values: {loaded_from_cif.is_hem_pocket[:3]}") |
| 211 | + |
| 212 | +# %% |
| 213 | + |
| 214 | +# Clean up temporary files |
| 215 | +import shutil |
| 216 | + |
| 217 | +shutil.rmtree(temp_dir) |
| 218 | +print(f"✓ Cleaned up temporary directory: {temp_dir}") |
| 219 | +print("✓ Successfully demonstrated structure annotation, saving, and loading!") |
| 220 | + |
| 221 | +######################################################################## |
| 222 | +# Related Examples |
| 223 | +# ---------- |
| 224 | +# |
| 225 | +# - :doc:`pocket_conditioning_transform` - Create custom transforms for ligand pocket identification and ML feature generation |
0 commit comments