rf2aa as a Python module

README.md

@@ -41,6 +41,12 @@ cd rf2aa/SE3Transformer/
 pip3 install --no-cache-dir -r requirements.txt
 python3 setup.py install
 cd ../../
+
+```
+Change the default checkpoint_path/database paths in `RoseTTAFold-All-Atom/rf2aa/config/inference/base.yaml` 
+then install `rf2aa` as a python module that can be called everywhere.
+```shell
+pip install . --no-dependencies   --no-cache-dir
 ```
 4. Configure signalp6 after downloading a licensed copy of it from https://services.healthtech.dtu.dk/services/SignalP-6.0/
 ```

install_dependencies.sh

@@ -20,3 +20,17 @@ echo "Downloading cs-blast ..."
 wget http://wwwuser.gwdg.de/~compbiol/data/csblast/releases/csblast-2.2.3_${platform}.tar.gz -O csblast-2.2.3.tar.gz
 mkdir -p csblast-2.2.3
 tar xf csblast-2.2.3.tar.gz -C csblast-2.2.3 --strip-components=1
+
+mv csblast-2.2.3 $CONDA_PREFIX/share/
+
+rm -f csblast-2.2.3.tar.gz
+
+# https://github.com/baker-laboratory/RoseTTAFold-All-Atom/issues/5#issuecomment-1990991606
+# https://github.com/RosettaCommons/RoseTTAFold/issues/13#issuecomment-1405850297
+echo "Downloading blast ..."
+wget https://ftp.ncbi.nlm.nih.gov/blast/executables/legacy.NOTSUPPORTED/2.2.26/blast-2.2.26-x64-linux.tar.gz
+mkdir -p blast-2.2.26
+tar -xf blast-2.2.26-x64-linux.tar.gz -C blast-2.2.26
+
+mv blast-2.2.26 $CONDA_PREFIX/share/
+rm -f blast-2.2.26-x64-linux.tar.gz

pyproject.toml

@@ -0,0 +1,22 @@
+[build-system]
+requires = ["poetry-core>=1.0.0,<2.0.0"]
+build-backend = "poetry.core.masonry.api"
+
+[tool.poetry]
+name = "rfaa"
+version = "1.0.0"
+description = "RoseTTAFold-All-Atom code"
+authors = ["Name <email@address>"]
+
+readme = "README.md"
+license = "MIT"
+repository = "https://github.com/baker-laboratory/RoseTTAFold-All-Atom"
+classifiers = [
+    "Topic :: Scientific/Engineering :: Biochemistry",
+    "Topic :: Scientific/Engineering :: Protein Engineering"
+]
+
+packages = [
+    { include = "rf2aa" },
+    { include = "rf2aa/*.py" },
+]

rf2aa/config/inference/base.yaml

@@ -1,12 +1,15 @@
 job_name: "structure_prediction"
-output_path: ""
-checkpoint_path: RFAA_paper_weights.pt
+output_path: "output/structure_prediction"
+checkpoint_path: /mnt/db/weights/RoseTTAFoldAA/RFAA_paper_weights.pt
 database_params:
-  sequencedb: ""
-  hhdb: "pdb100_2021Mar03/pdb100_2021Mar03"
+  DB_UR30: "/mnt/db/uniref30_uc30/UniRef30_2022_02/UniRef30_2022_02"
+  DB_BFD: "/mnt/db/bfd/bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt"
+  DB_PDB100: "/mnt/db/pdb100/pdb100_2021Mar03/pdb100_2021Mar03"
   command: make_msa.sh
   num_cpus: 4
   mem: 64
+
+force_cpu: False # force to use cpu if CUDA out of memory
 protein_inputs: null
 na_inputs: null
 sm_inputs: null

rf2aa/data/preprocessing.py

@@ -4,6 +4,8 @@
 import subprocess
 
 #from rf2aa.run_inference import ModelRunner
+script_dir=os.path.dirname(os.path.abspath(__file__)) #RoseTTAFold-All-Atom/rf2aa/data
+msa_script_dir=os.path.abspath(os.path.join(script_dir, '..','input_prep'))   #RoseTTAFold-All-Atom/rf2aa/input_prep
 
 
 def make_msa(
@@ -17,19 +19,25 @@ def make_msa(
     out_dir.mkdir(parents=True, exist_ok=True)
 
     command = model_runner.config.database_params.command
-    search_base = model_runner.config.database_params.sequencedb
+
+# sequence databases
+    DB_UR30=model_runner.config.database_params.DB_UR30
+    DB_BFD=model_runner.config.database_params.DB_BFD
     num_cpus = model_runner.config.database_params.num_cpus
     ram_gb = model_runner.config.database_params.mem
-    template_database = model_runner.config.database_params.hhdb
+    template_database = model_runner.config.database_params.DB_PDB100
 
     out_a3m = out_dir / "t000_.msa0.a3m"
     out_atab = out_dir / "t000_.atab"
     out_hhr = out_dir / "t000_.hhr"
     if out_a3m.exists() and out_atab.exists() and out_hhr.exists():
         return out_a3m, out_hhr, out_atab
 
-    search_command = f"./{command} {fasta_file} {out_dir} {num_cpus} {ram_gb} {search_base} {template_database}"
+    search_command = f"{msa_script_dir}/{command} {os.path.abspath(fasta_file)} {os.path.abspath(out_dir)} {num_cpus} {ram_gb} {DB_UR30} {DB_BFD} {template_database}"
     print(search_command)
     _ = subprocess.run(search_command, shell=True)
+
+    if _.returncode != 0:
+        raise RuntimeError(f"Failed to execute command {search_command}")
     return out_a3m, out_hhr, out_atab
 

make_msa.sh → rf2aa/input_prep/make_msa.sh

@@ -1,4 +1,5 @@
 #!/bin/bash
+set -e
 
 # inputs
 in_fasta="$1"
@@ -8,17 +9,23 @@ out_dir="$2"
 CPU="$3"
 MEM="$4"
 
+export BLAST_LEGACY=$CONDA_PREFIX/share/blast-2.2.26/blast-2.2.26
+export BLASTMAT=$BLAST_LEGACY/data/
+export BLAST_LEGACY_BIN=$BLAST_LEGACY/bin
+
+export PATH=$BLAST_LEGACY_BIN:$PATH
+
+# sequence databases
+DB_UR30="$5"
+DB_BFD="$6"
+
 # template database
-DB_TEMPL="$5"
+DB_TEMPL="$7"
 
 # current script directory (i.e., pipe directory)
 SCRIPT=`realpath -s $0`
 export PIPE_DIR=`dirname $SCRIPT`
 
-# sequence databases
-DB_UR30="$PIPE_DIR/uniclust/UniRef30_2021_06"
-DB_BFD="$PIPE_DIR/bfd/bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt"
-
 # Running signalP 6.0
 mkdir -p $out_dir/signalp
 tmp_dir="$out_dir/signalp"
@@ -30,7 +37,7 @@ then
 fi
 
 # setup hhblits command
-export HHLIB=/software/hhsuite/build/bin/
+export HHLIB=$(dirname $(which hhblits))
 export PATH=$HHLIB:$PATH
 HHBLITS_UR30="hhblits -o /dev/null -mact 0.35 -maxfilt 100000000 -neffmax 20 -cov 25 -cpu $CPU -nodiff -realign_max 100000000 -maxseq 1000000 -maxmem $MEM -n 4 -d $DB_UR30"
 HHBLITS_BFD="hhblits -o /dev/null -mact 0.35 -maxfilt 100000000 -neffmax 20 -cov 25 -cpu $CPU -nodiff -realign_max 100000000 -maxseq 1000000 -maxmem $MEM -n 4 -d $DB_BFD"
@@ -113,7 +120,7 @@ fi
 
 echo "Running PSIPRED"
 mkdir -p $out_dir/log
-$PIPE_DIR/input_prep/make_ss.sh $out_dir/t000_.msa0.a3m $out_dir/t000_.ss2 > $out_dir/log/make_ss.stdout 2> $out_dir/log/make_ss.stderr
+$PIPE_DIR/make_ss.sh $out_dir/t000_.msa0.a3m $out_dir/t000_.ss2 > $out_dir/log/make_ss.stdout 2> $out_dir/log/make_ss.stderr
 
 if [ ! -s $out_dir/t000_.hhr ]
 then

input_prep/make_ss.sh → rf2aa/input_prep/make_ss.sh

@@ -1,21 +1,25 @@
 #!/bin/bash
 # From: https://github.com/RosettaCommons/RoseTTAFold
 
+set -e
+
 DATADIR="$CONDA_PREFIX/share/psipred_4.01/data"
+csblast_dir="$CONDA_PREFIX/share/csblast-2.2.3"
 echo $DATADIR
 
 i_a3m="$1"
 o_ss="$2"
 
 ID=$(basename $i_a3m .a3m).tmp
 
-$PIPE_DIR/csblast-2.2.3/bin/csbuild -i $i_a3m -I a3m -D $PIPE_DIR/csblast-2.2.3/data/K4000.crf -o $ID.chk -O chk
+$csblast_dir/bin/csbuild -i $i_a3m -I a3m -D $csblast_dir/data/K4000.crf -o $ID.chk -O chk
 
 head -n 2 $i_a3m > $ID.fasta
 echo $ID.chk > $ID.pn
 echo $ID.fasta > $ID.sn
 
 makemat -P $ID
+echo "psipred $ID.mtx $DATADIR/weights.dat $DATADIR/weights.dat2 $DATADIR/weights.dat3 > $ID.ss"
 psipred $ID.mtx $DATADIR/weights.dat $DATADIR/weights.dat2 $DATADIR/weights.dat3 > $ID.ss
 psipass2 $DATADIR/weights_p2.dat 1 1.0 1.0 $i_a3m.csb.hhblits.ss2 $ID.ss > $ID.horiz
 

rf2aa/run_inference.py

@@ -24,9 +24,9 @@ def __init__(self, config) -> None:
         self.config = config
         initialize_chemdata(self.config.chem_params)
         FFindexDB = namedtuple("FFindexDB", "index, data")
-        self.ffdb = FFindexDB(read_index(config.database_params.hhdb+'_pdb.ffindex'),
-                              read_data(config.database_params.hhdb+'_pdb.ffdata'))
-        self.device = "cuda:0" if torch.cuda.is_available() else "cpu"
+        self.ffdb = FFindexDB(read_index(config.database_params.DB_PDB100+'_pdb.ffindex'),
+                              read_data(config.database_params.DB_PDB100+'_pdb.ffdata'))
+        self.device = "cuda:0" if torch.cuda.is_available() and not config.force_cpu else "cpu"
         self.xyz_converter = XYZConverter()
         self.deterministic = config.get("deterministic", False)
         self.molecule_db = load_pdb_ideal_sdf_strings()