def compute_similarity(mref, molecules, fp_type="DL"):
fp_ref = fps.fingerprint(mref, fp_type)
for i, m in enumerate(molecules):
if m:
yield ml.molec_properties(m, fp_ref=fp_ref, fp_type=fp_type)
else:
print("Molecule {}".format(i))
def combi_substructure_search(sdfs, molecules_db):
print("Searching for substructure")
for i, m in tqdm(enumerate(molecules_db)):
if not m:
print("Skipping {}".format(i))
continue
substructs_found = [False] * len(sdfs)
for i, sdf in enumerate(sdfs):
for m_ref in Chem.SDMolSupplier(sdf):
if m.HasSubstructMatch(m_ref, useChirality=True):
substructs_found[i] = True
if all(substructs_found):
yield ml.molec_properties(m)
def search_substructure(molecule_query, molecules_db):
print("Searching for substructure")
all_substructs_found = True
for i, m in tqdm(enumerate(molecules_db)):
if not m:
print("Skipping {}".format(i))
continue
all_substructs_found = True
for m_ref in molecule_query:
if not m.HasSubstructMatch(m_ref, useChirality=True):
all_substructs_found = False
if all_substructs_found:
yield ml.molec_properties(m)
def compute_similarity_several_mols(mrefs, molecules, fp_type="DL"):
fp_refs = [
fps.fingerprint(mref, fp_type=fp_type) for mref in mrefs if mref
]
for m in molecules:
if m:
molec = ml.molec_properties(m, fp_type=fp_type)
molec.similarities = [
DataStructs.FingerprintSimilarity(fp_ref, molec.fp)
for fp_ref in fp_refs
]
yield molec
else:
pass
def compute_similarity_severl_fp(
mref,
molecules,
fp_types=["DL", "circular", "torsions", "MACCS"],
tresholds=[0.7, 0.4, 0.7, 0.4]):
for i, m in tqdm(enumerate(molecules)):
chosen = False
for j, fp_type in enumerate(fp_types):
fp_ref = fps.fingerprint(mref, fp_type=fp_type)
if not chosen:
if m:
molec = ml.molec_properties(m,
fp_ref=fp_ref,
fp_type=fp_type)
if molec.similarity > float(tresholds[j]):
chosen = True
yield molec
else:
print("Molecule {}".format(i))
else:
break
def main(molecule_query,
molecules_db,
n_bins=100,
fp_types=["DL", "MACCS", "circular"],
test=None,
dim_type="umap"):
#Analysis code
for fp_type in fp_types:
# Retrieve fingerprints and similarity for analysis
counts = []
svg = [None] * 100
fp = [None] * 100
similarity = [0] * 100
for m in tqdm(
sim.compute_similarity(molecule_query, molecules_db, fp_type)):
#Build similarity counts
counts_step, bins = pt.histogram(m.similarity, n_bins=n_bins)
counts = np.add(counts,
counts_step) if len(counts) != 0 else counts_step
#Build fingeprints
idx = np.argmin(similarity)
less_similar = similarity[idx]
if m.similarity > less_similar:
fp[idx] = m.fp
similarity[idx] = m.similarity
svg[idx] = m.to_svg()
#Plot similarity distribution
if not test:
fig, ax = plt.subplots()
ax.hist(bins[1:], bins, weights=counts)
plt.savefig("similarity_hist_{}.png".format(fp_type))
#Retrieve similarity distribution data
values = np.vstack([bins[1:], counts]).T
np.savetxt("counts_{}.txt".format(fp_type), values, fmt=['%f', '%d'])
if fp_type == "circular":
#Add reference molecule
mref = ml.molec_properties(molecule_query, fp_type=fp_type)
fp.insert(0, mref.fp)
similarity.insert(0, 1)
svg.insert(0, mref.to_svg())
#Clean results
svg = [value for value in svg if value]
fp = [value for value in fp if value]
similarity = [value for value in similarity if value != 0]
#Reduce Dimension
X = np.asarray([fp2arr(f) for f in fp])
embedding = dm.ReduceDimension(X, 2).run(dim_type)
#Plot Reduce Dimension graph
if not test:
pt.interactive_map(embedding[:, 0],
embedding[:, 1],
svg,
color=np.array(similarity))