def GetDistanceMatrix(data, metric, isSimilarity=1):
""" data should be a list of tuples with fingerprints in position 1
(the rest of the elements of the tuple are not important)
Returns the symmetric distance matrix
(see ML.Cluster.Resemblance for layout documentation)
"""
nPts = len(data)
res = numpy.zeros((nPts * (nPts - 1) / 2), numpy.float)
nSoFar = 0
for col in xrange(1, nPts):
for row in xrange(col):
fp1 = data[col][1]
fp2 = data[row][1]
if fp1.GetNumBits() > fp2.GetNumBits():
fp1 = DataStructs.FoldFingerprint(
fp1,
fp1.GetNumBits() / fp2.GetNumBits())
elif fp2.GetNumBits() > fp1.GetNumBits():
fp2 = DataStructs.FoldFingerprint(
fp2,
fp2.GetNumBits() / fp1.GetNumBits())
sim = metric(fp1, fp2)
if isSimilarity:
sim = 1. - sim
res[nSoFar] = sim
nSoFar += 1
return res
def GetDistanceMatrix(data, metric, isSimilarity=1):
""" data should be a list of tuples with fingerprints in position 1
(the rest of the elements of the tuple are not important)
Returns the symmetric distance matrix
(see ML.Cluster.Resemblance for layout documentation)
"""
nPts = len(data)
distsMatrix = numpy.zeros((nPts * (nPts - 1) // 2), dtype=numpy.float64)
nSoFar = 0
for col in range(1, nPts):
fp1 = data[col][1]
nBits1 = fp1.GetNumBits()
for row in range(col):
fp2 = data[row][1]
nBits2 = fp2.GetNumBits()
if nBits1 > nBits2:
fp1 = DataStructs.FoldFingerprint(fp1, nBits1 / nBits2)
elif nBits2 > nBits1:
fp2 = DataStructs.FoldFingerprint(fp2, nBits2 / nBits1)
if isSimilarity:
distsMatrix[nSoFar] = 1.0 - metric(fp1, fp2)
else:
distsMatrix[nSoFar] = metric(fp1, fp2)
nSoFar += 1
return distsMatrix
def test6(self):
""" check that the bits in a signature of size N which has been folded in half
are the same as those in a signature of size N/2
"""
smis = ['CCC(O)C(=O)O', 'c1ccccc1', 'C1CCCCC1', 'C1NCCCC1', 'CNCNCNC']
for smi in smis:
m = Chem.MolFromSmiles(smi)
fp1 = Chem.RDKFingerprint(m, 2, 7, 4096)
fp2 = DataStructs.FoldFingerprint(fp1, 2)
fp3 = Chem.RDKFingerprint(m, 2, 7, 2048)
assert tuple(fp2.GetOnBits()) == tuple(fp3.GetOnBits())
fp2 = DataStructs.FoldFingerprint(fp2, 2)
fp3 = Chem.RDKFingerprint(m, 2, 7, 1024)
assert tuple(fp2.GetOnBits()) == tuple(fp3.GetOnBits())
fp2 = DataStructs.FoldFingerprint(fp1, 4)
assert tuple(fp2.GetOnBits()) == tuple(fp3.GetOnBits())
def fingerprint_reactions(reactions, fp_dim):
fps = []
for r in reactions:
rxn = AllChem.ReactionFromSmarts(r)
fp = AllChem.CreateStructuralFingerprintForReaction(rxn)
fold_factor = fp.GetNumBits() // fp_dim
fp = DataStructs.FoldFingerprint(fp, fold_factor)
fps.append(fp)
return fps
def FoldFingerprintToTargetDensity(fp, **fpArgs):
nOn = fp.GetNumOnBits()
nTot = fp.GetNumBits()
while (float(nOn) / nTot < fpArgs['tgtDensity']):
if nTot / 2 > fpArgs['minSize']:
fp = DataStructs.FoldFingerprint(fp, 2)
nOn = fp.GetNumOnBits()
nTot = fp.GetNumBits()
else:
break
return fp
def test15FoldFingerprint(self):
for cls in [DataStructs.ExplicitBitVect, DataStructs.SparseBitVect]:
fp = cls(8)
fp[0] = 1
fp[1] = 1
fp[6] = 1
ffp = DataStructs.FoldFingerprint(fp)
self.assertTrue(ffp[0])
self.assertTrue(ffp[1])
self.assertTrue(ffp[2])
self.assertFalse(ffp[3])
def SMILE2Matrix(smile_list):
# To ECFP6
def ToECFP(id_smile):
cid = id_smile[0]
smile = id_smile[1]
mol = Chem.MolFromSmiles(smile)
return [cid, AllChem.GetMorganFingerprintAsBitVect(mol, 3, nBits=1024)]
fps = map(ToECFP, smile_list)
np_fps = []
ids = []
for fp in fps:
arr = np.zeros((1, ))
vfp = DataStructs.FoldFingerprint(fp[1], 4)
DataStructs.ConvertToNumpyArray(vfp, arr)
ids.append(fp[0])
np_fps.append(arr)
return ids, np_fps
