• FDA-approved small drug 3D structure dataset
  • 1600 FDA-approved small molecule drugs
  • Molecule sizes
    • average heavy atom numbers and standard deviation
  • 3D structures of the chemicals
    • Drug Bank id to pdb id
    • Filter the unbound proteins
    • Cluster the bound proteins based on sequence similarity

FDA-approved small drug 3D structure dataset

1600 FDA-approved small molecule drugs

1. http://www.drugbank.ca/downloads#structures
2. ./dat/approved.txt

Molecule sizes

average heavy atom numbers and standard deviation

1. script

   ./script/count_atms.py

2. result

   ./dat/approved_heavy_atom_num.txt ./dat/elements.txt

DONE 3D structures of the chemicals

DONE Drug Bank id to pdb id

1. pdb –> uniprt

   ./dat/pdbsws_chain.txt

2. drug bank id –> uniprt ids

   1. ./script/crawler.py
   2. ./dat/drugid_uniprturls.dat

3. uniprt id –> pdb id

   1. ./script/uniprt_pdb.py
   2. ./dat/uniprt_to_pdb.dat

4. drug bank id –> pdb id

   1. ./script/query_pdb.py
   2. ./dat/drugid_pdb.dat
   3. 82773 pdbs
   4. 138078 chains

5. table

   import pickle
   import pandas as pd
   
   dat_ifn = "./dat/drugid_pdb.dat"
   
   with open(dat_ifn, 'r') as f:
       dat = pickle.load(f)
   
   dset = pd.DataFrame({key: list(val) for key, val in dat.iteritems()})

DONE Filter the unbound proteins

1. DONE Extract the ligand atoms from the CIF files

   1. ./script/qmol_CIF2PDB.pl (not compatible with latest perl modules)
   2. work/jaydy/dat/fda_pdb_mb

2. DONE Cluster drug ligands

   1. cmd

      $ cd ~/Workspace/Bitbucket/fda-approved-drugs/dat/
      $ python /home/jaydy/Workspace/Bitbucket/geauxcluster/src/geauxcluster.py -i approved.txt -o approved.txt.clusters -k DRUGBANK_ID

   2. script ./script/clustering.py

3. DONE Compare the pdb ligand with drug bank ligand

   1. calculate tanimoto coefficient using Kcombu
      1. ./script/calculateTanimota.py
      2. pkcombu Segmentation fault for
         1. DB01049
         2. to check pkcombu -A /work/jaydy/working/kcombu_run/DB01049_fda.sdf -B /work/jaydy/dat/fda_pdb_mb/ib/3ibdA.LG3.pdb
         3. DB00707
   2. refuse proteins if their ligand's tanimoto < 0.9
      1. ./script/calculateTanimoto.py
      2. TANI_DIR = "/work/jaydy/working/tanimoto"

TODO Cluster the bound proteins based on sequence similarity

1. DONE add missing atoms

   ./script/fixPrt.py

2. DONE clean the proteins sequences

   python ./script/pdb_to_fasta.py > run.sh sh ./script/run.sh

   1. ctrip, heavy atom model
      1. /project/michal/apps/jackal_64bit/bin/ctrip
   2. ./script/pdb2pdb.pl move pdb sequence to start at 1
   3. ./script/pdb2fasta.pl convert pdb to fasta
   4. ./script/fasta2fasta.pl break the lines at 80

3. DONE cd-hit to cluster the sequences

   1. do not count the prt with seq length > 600
   2. cluster python ./script/cluster_seq.py > ./script/cluster_seq.sh cd /work/jaydy/working/cluster sh /home/jaydy/Workspace/Bitbucket/fda-approved-drugs/script/cluster_seq.sh
   3. collect /home/jaydy/Workspace/Bitbucket/fda-approved-drugs/script/collect.py

4. TODO

   1. original data set of 1554 fda-approved drugs

      1. script ./script/count_atms.py ./dat/drug_size.txt

      2. result

         count	1554.000000
         mean	26.257400
         std	18.172654
         min	1.000000
         25%	18.000000
         50%	23.000000
         75%	30.000000
         max	419.000000

      3. one atom drugs

         DrugBank ID	Chem
         DB01356	Li
         DB01370	Al
         DB01592	Fe
         DB01593	Zn

      4. 2 σ range (0, 62.6)

      5. 1 σ range (8, 44)

   2. after processing

      1. script ./script/drug_size.py

      2. result

         statistics	#HeavyAtom
         count	197.000000
         mean	23.928934
         std	12.470170
         min	6.000000
         25%	16.000000
         50%	21.000000
         75%	29.000000
         max	93.000000

      3. within the range of (8, 44)

         1. dat ./dat/representative_drugs.csv

            stats	#HeavyAtom
            count	186.000000
            mean	22.639785
            std	8.456204
            min	8.000000
            25%	16.000000
            50%	21.000000
            75%	29.000000
            max	44.000000

         2. 3D structures

            1. protein bound ligands

               import itertools
               import shutil
               import os
               import pandas as pd
               import cPickle
               
               represents_ifn = "./dat/representative_drugs.csv"
               dat_ifn = "./dat/dat.pkl"
               
               with open(dat_ifn, 'r') as f:
                   dset = cPickle.load(f)
               
               represents = pd.read_csv(represents_ifn, index_col=0)
               
               inter = dset[dset.DRUGBANK_ID.isin(represents.id)]
               inter.to_csv("./dat/fda_drugs_prt_lig.csv")
               
               drugs = {}
               for index, row in inter.iterrows():
                   drugs[row.DRUGBANK_ID] = row.ProteinBoundLig.split()
               
               complexes = set(itertools.chain(*drugs.values()))
               print "# complexes %d" % (len(complexes))
               
               proteins = set([complex_id.split('.')[0][:4] for complex_id in complexes])
               print "# proteins %d" % (len(proteins))
               
               # retrive the 3D structures
               DAT_DIR = "/work/jaydy/dat/fda_pdb_mb"
               STRUCTURE_DIR = "/work/jaydy/dat/fda_drug_structures"
               
               def getPaths(ligand_id):
                   mid_two = ligand_id[1:3]
                   ligand_path = os.path.join(DAT_DIR, mid_two, ligand_id)
                   prt_id = ligand_id.split('.')[0] + '.pdb'
                   prt_path = os.path.join(DAT_DIR, mid_two, prt_id)
                   return ligand_path, prt_path
               
               def copyFiles(drug_id):
                   dat_dir = os.path.join(STRUCTURE_DIR, drug_id)
                   os.makedirs(dat_dir)
                   for ligand_id in drugs[drug_id]:
                       ligand_path, prt_path = getPaths(ligand_id)
                       shutil.copy(ligand_path, dat_dir)
                       shutil.copy(prt_path, dat_dir)
               
                   print drug_id, "done"
               
               
               for drug_id in drugs:
                   copyFiles(drug_id)