dependency update openmm 7.5.1 -> 7.7.0 (#61)

README.md

@@ -39,7 +39,6 @@ git clone https://github.com/hpcaitech/FastFold
 cd FastFold
 conda env create --name=fastfold -f environment.yml
 conda activate fastfold
-bash scripts/patch_openmm.sh
 python setup.py install
 ```
 

docker/Dockerfile

@@ -5,7 +5,7 @@ FROM hpcaitech/cuda-conda:11.3
 RUN yum install -y patch
 
 RUN conda install pytorch==1.10.0 torchvision torchaudio cudatoolkit=11.3 -c pytorch \
- && conda install setuptools=59.5.0 openmm=7.5.1 pdbfixer -c conda-forge \
+ && conda install setuptools=59.5.0 openmm=7.7.0 pdbfixer -c conda-forge \
  && conda install hmmer==3.3.2 hhsuite=3.3.0 kalign2=2.04 -c bioconda
 
 RUN pip install biopython==1.79 dm-tree==0.1.6 ml-collections==0.1.0 numpy==1.21.2 \
@@ -17,5 +17,4 @@ RUN pip install colossalai==0.1.8+torch1.10cu11.3 -f https://release.colossalai.
 # prepare environment
 Run git clone https://github.com/hpcaitech/FastFold.git\
  && cd ./FastFold \
- && /bin/bash scripts/patch_openmm.sh \
  && python setup.py install

environment.yml

@@ -23,7 +23,7 @@ dependencies:
   - conda-forge::python=3.8
   - conda-forge::setuptools=59.5.0
   - conda-forge::pip
-  - conda-forge::openmm=7.5.1
+  - conda-forge::openmm=7.7.0
   - conda-forge::pdbfixer
   - bioconda::hmmer==3.3.2
   - bioconda::hhsuite==3.3.0

fastfold/relax/amber_minimize.py

@@ -27,10 +27,10 @@ import fastfold.model.loss as loss
 from fastfold.relax import cleanup, utils
 import ml_collections
 import numpy as np
-from simtk import openmm
-from simtk import unit
-from simtk.openmm import app as openmm_app
-from simtk.openmm.app.internal.pdbstructure import PdbStructure
+import openmm
+from openmm import unit
+from openmm import app as openmm_app
+from openmm.app.internal.pdbstructure import PdbStructure
 
 ENERGY = unit.kilocalories_per_mole
 LENGTH = unit.angstroms

fastfold/relax/cleanup.py

@@ -20,8 +20,8 @@ cases like removing chains of length one (see clean_structure).
 import io
 
 import pdbfixer
-from simtk.openmm import app
-from simtk.openmm.app import element
+from openmm import app
+from openmm.app import element
 
 
 def fix_pdb(pdbfile, alterations_info):

fastfold/relax/utils.py

@@ -18,8 +18,8 @@ import collections
 from fastfold.common import residue_constants
 from Bio import PDB
 import numpy as np
-from simtk.openmm import app as openmm_app
-from simtk.openmm.app.internal.pdbstructure import PdbStructure
+from openmm import app as openmm_app
+from openmm.app.internal.pdbstructure import PdbStructure
 
 
 def overwrite_pdb_coordinates(pdb_str: str, pos) -> str:

inference.py

@@ -280,7 +280,7 @@ def inference_monomer_model(args):
     seqs, tags = parse_fasta(fasta)
 
     for tag, seq in zip(tags, seqs):
-        print(f"tag:{tag} seq:{seq}")
+        print(f"tag:{tag}\nseq[{len(seq)}]:{seq}")
         batch = [None]
         
         fasta_path = os.path.join(args.output_dir, "tmp.fasta")

inference.sh

-python inference.py target.fasta /data/pdb_mmcif/mmcif_files \
+# add `--gpus [N]` to use N gpus for inference
+# add `--enable_workflow` to use parallel workflow for data processing
+# add `--use_precomputed_alignments [path_to_alignments]` to use precomputed msa
+
+python inference.py target.fasta data/pdb_mmcif/mmcif_files \
     --output_dir ./ \
-    --gpus 2 \
     --uniref90_database_path data/uniref90/uniref90.fasta \
     --mgnify_database_path data/mgnify/mgy_clusters_2018_12.fa \
     --pdb70_database_path data/pdb70/pdb70 \
@@ -9,5 +12,4 @@ python inference.py target.fasta /data/pdb_mmcif/mmcif_files \
     --jackhmmer_binary_path `which jackhmmer` \
     --hhblits_binary_path `which hhblits` \
     --hhsearch_binary_path `which hhsearch` \
-    --kalign_binary_path `which kalign`  \
-    # --enable_workflow
\ No newline at end of file
+    --kalign_binary_path `which kalign`

scripts/openmm.patch

-Index: simtk/openmm/app/topology.py
-===================================================================
---- simtk.orig/openmm/app/topology.py
-+++ simtk/openmm/app/topology.py
-@@ -356,19 +356,35 @@
-         def isCyx(res):
-             names = [atom.name for atom in res._atoms]
-             return 'SG' in names and 'HG' not in names
-+        # This function is used to prevent multiple di-sulfide bonds from being
-+        # assigned to a given atom. This is a DeepMind modification.
-+        def isDisulfideBonded(atom):
-+            for b in self._bonds:
-+                if (atom in b and b[0].name == 'SG' and
-+                    b[1].name == 'SG'):
-+                    return True
-+
-+            return False
- 
-         cyx = [res for res in self.residues() if res.name == 'CYS' and isCyx(res)]
-         atomNames = [[atom.name for atom in res._atoms] for res in cyx]
-         for i in range(len(cyx)):
-             sg1 = cyx[i]._atoms[atomNames[i].index('SG')]
-             pos1 = positions[sg1.index]
-+            candidate_distance, candidate_atom = 0.3*nanometers, None
-             for j in range(i):
-                 sg2 = cyx[j]._atoms[atomNames[j].index('SG')]
-                 pos2 = positions[sg2.index]
-                 delta = [x-y for (x,y) in zip(pos1, pos2)]
-                 distance = sqrt(delta[0]*delta[0] + delta[1]*delta[1] + delta[2]*delta[2])
--                if distance < 0.3*nanometers:
--                    self.addBond(sg1, sg2)
-+                if distance < candidate_distance and not isDisulfideBonded(sg2):
-+                    candidate_distance = distance
-+                    candidate_atom = sg2
-+            # Assign bond to closest pair.
-+            if candidate_atom:
-+                self.addBond(sg1, candidate_atom)
-+
-+
- 
- class Chain(object):
-     """A Chain object represents a chain within a Topology."""

scripts/patch_openmm.sh

-#!/bin/bash
-
-PYTHON_SITE_PATH=$(python3 -c 'import sysconfig; print(sysconfig.get_paths()["purelib"])')
-PATCH_PATH=$(realpath $(dirname $0))
-
-pushd ${PYTHON_SITE_PATH} \
-	&& patch -p0 < ${PATCH_PATH}/openmm.patch \
-	&& popd