Merge branch 'main' into multimer

Dockerfile

@@ -31,6 +31,6 @@ COPY setup.py /opt/openfold/setup.py
 COPY lib/openmm.patch /opt/openfold/lib/openmm.patch
 RUN wget -q -P /opt/openfold/openfold/resources \
     https://git.scicore.unibas.ch/schwede/openstructure/-/raw/7102c63615b64735c4941278d92b554ec94415f8/modules/mol/alg/src/stereo_chemical_props.txt
-RUN patch -p0 -d /opt/conda/lib/python3.7/site-packages/ < /opt/openfold/lib/openmm.patch
+RUN patch -p0 -d /opt/conda/lib/python3.9/site-packages/ < /opt/openfold/lib/openmm.patch
 WORKDIR /opt/openfold
 RUN python3 setup.py install

README.md

@@ -298,7 +298,7 @@ python3 scripts/generate_chain_data_cache.py \
 ```
 
 where the `cluster_file` argument is a file of chain clusters, one cluster
-per line (e.g. [PDB40](https://cdn.rcsb.org/resources/sequence/clusters/clusters-by-entity-40.txt)).
+per line.
 
 Optionally, download an AlphaFold-style validation set from 
 [CAMEO](https://cameo3d.org) using `scripts/download_cameo.py`. Use the 

environment.yml

@@ -4,7 +4,7 @@ channels:
   - bioconda
   - pytorch
 dependencies:
-  - conda-forge::python=3.7
+  - conda-forge::python=3.9
   - conda-forge::setuptools=59.5.0
   - conda-forge::pip
   - conda-forge::openmm=7.5.1

notebooks/OpenFold.ipynb

@@ -97,6 +97,7 @@
         "#@markdown **Note**: This installs the software on the Colab \n",
         "#@markdown notebook in the cloud and not on your computer.\n",
         "\n",
+        "import sys\n",
         "from IPython.utils import io\n",
         "import os\n",
         "import subprocess\n",
@@ -104,6 +105,8 @@
         "\n",
         "TQDM_BAR_FORMAT = '{l_bar}{bar}| {n_fmt}/{total_fmt} [elapsed: {elapsed} remaining: {remaining}]'\n",
         "\n",
+        "python_version = '.'.join(sys.version.split('.')[:2]) #get string like \"3.9\"\n",
+        "\n",
         "try:\n",
         "  with io.capture_output() as captured:\n",
         "    %shell sudo apt install --quiet --yes hmmer\n",
@@ -125,12 +128,15 @@
         "    %shell conda install -y -q -c conda-forge -c bioconda \\\n",
         "      kalign2=2.04 \\\n",
         "      hhsuite=3.3.0 \\\n",
-        "      python=3.8 \\\n",
+        "      python={python_version} \\\n",
+        "      openmm=7.7.0 \\\n",
+        "      pdbfixer \\\n",
         "      2>&1 1>/dev/null\n",
         "    %shell pip install -q \\\n",
         "      ml-collections==0.1.0 \\\n",
         "      PyYAML==5.4.1 \\\n",
-        "      biopython==1.79\n",
+        "      biopython==1.79 \\\n",
+        "      modelcif==0.7\n",
         "\n",
         "    # Create a ramdisk to store a database chunk to make Jackhmmer run fast.\n",
         "    %shell sudo mkdir -m 777 --parents /tmp/ramdisk\n",
@@ -181,13 +187,6 @@
         "    %shell cp -f /content/stereo_chemical_props.txt /content/openfold/openfold/resources\n",
         "    %shell /usr/bin/python3 -m pip install -q ./openfold\n",
         "\n",
-        "    %shell conda install -y -q -c conda-forge openmm=7.5.1\n",
-        "    # Apply OpenMM patch.\n",
-        "    %shell pushd /opt/conda/lib/python3.8/site-packages/ && \\\n",
-        "        patch -p0 < /content/openfold/lib/openmm.patch && \\\n",
-        "        popd\n",
-        "    %shell conda install -y -q -c conda-forge pdbfixer=1.7\n",
-        "\n",
         "    if(weight_set == 'AlphaFold'):\n",
         "      %shell mkdir --parents \"{ALPHAFOLD_PARAMS_DIR}\"\n",
         "      %shell wget -O {ALPHAFOLD_PARAMS_PATH} {ALPHAFOLD_PARAM_SOURCE_URL}\n",
@@ -220,8 +219,8 @@
         "import unittest.mock\n",
         "import sys\n",
         "\n",
-        "sys.path.insert(0, '/usr/local/lib/python3.8/site-packages/')\n",
-        "sys.path.append('/opt/conda/lib/python3.8/site-packages')\n",
+        "sys.path.insert(0, f'/usr/local/lib/python{python_version}/site-packages/')\n",
+        "sys.path.append(f'/opt/conda/lib/python{python_version}/site-packages')\n",
         "\n",
         "# Allows us to skip installing these packages\n",
         "unnecessary_modules = [\n",
@@ -791,4 +790,4 @@
       ]
     }
   ]
-}
+}

openfold/model/msa.py

@@ -457,9 +457,6 @@ class MSAColumnGlobalAttention(nn.Module):
         m = m.transpose(-2, -3)
         mask = mask.transpose(-1, -2)
 
-        # [*, N_res, N_seq, C_in]
-        #m = self.layer_norm_m(m)
-
         if chunk_size is not None:
             m = self._chunk(m, mask, chunk_size, use_lma=use_lma) 
         else:

openfold/model/triangular_multiplicative_update.py

@@ -433,8 +433,11 @@ class TriangleMultiplicativeUpdate(BaseTriangleMultiplicativeUpdate):
 
         # Prevents overflow of torch.matmul in combine projections in
         # reduced-precision modes
-        a = a / a.std()
-        b = b / b.std()
+        a_std = a.std()
+        b_std = b.std()
+        if(a_std != 0. and b_std != 0.):
+            a = a / a.std()
+            b = b / b.std()
 
         if(is_fp16_enabled()):
             with torch.cuda.amp.autocast(enabled=False):

openfold/np/relax/relax.py

@@ -57,7 +57,7 @@ class AmberRelaxation(object):
         self._use_gpu = use_gpu
 
     def process(
-        self, *, prot: protein.Protein, cif_output: bool
+        self, *, prot: protein.Protein, cif_output: bool = False
     ) -> Tuple[str, Dict[str, Any], np.ndarray]:
         """Runs Amber relax on a prediction, adds hydrogens, returns PDB string."""
         out = amber_minimize.run_pipeline(

openfold/np/residue_constants.py

@@ -409,7 +409,9 @@ residue_atoms = {
 # (The LDDT paper lists 7 amino acids as ambiguous, but the naming ambiguities
 # in LEU, VAL and ARG can be resolved by using the 3d constellations of
 # the 'ambiguous' atoms and their neighbours)
-# TODO: ^ interpret this
+# Because for LEU, VAL and ARG, no ambiguous exist when the prediction output is chi angle instead of the location of individual atoms.
+# For the rest, ASP and others, when you rotate the bond 180 degree, you get the same configuraiton due to symmetry.
+
 residue_atom_renaming_swaps = {
     "ASP": {"OD1": "OD2"},
     "GLU": {"OE1": "OE2"},

openfold/utils/script_utils.py

@@ -228,7 +228,7 @@ def prep_output(out, batch, feature_dict, feature_processor, config_preset, mult
     return unrelaxed_protein
 
 
-def relax_protein(config, model_device, unrelaxed_protein, output_directory, output_name, cif_output):
+def relax_protein(config, model_device, unrelaxed_protein, output_directory, output_name, cif_output=False):
     amber_relaxer = relax.AmberRelaxation(
         use_gpu=(model_device != "cpu"),
         **config.relax,
@@ -257,4 +257,4 @@ def relax_protein(config, model_device, unrelaxed_protein, output_directory, out
     with open(relaxed_output_path, 'w') as fp:
         fp.write(struct_str)
 
-    logger.info(f"Relaxed output written to {relaxed_output_path}...")
+    logger.info(f"Relaxed output written to {relaxed_output_path}...")

scripts/install_third_party_dependencies.sh

@@ -27,7 +27,7 @@ cd $CUR_DIR
 
 # Install DeepMind's OpenMM patch
 OPENFOLD_DIR=$PWD
-pushd lib/conda/envs/$ENV_NAME/lib/python3.7/site-packages/ \
+pushd lib/conda/envs/$ENV_NAME/lib/python3.9/site-packages/ \
     && patch -p0 < $OPENFOLD_DIR/lib/openmm.patch \
     && popd
 

setup.py

@@ -130,7 +130,7 @@ setup(
     classifiers=[
         'License :: OSI Approved :: Apache Software License',
         'Operating System :: POSIX :: Linux',
-        'Programming Language :: Python :: 3.7,'
+        'Programming Language :: Python :: 3.9,'
         'Topic :: Scientific/Engineering :: Artificial Intelligence',
     ],
 )