Spaces:
Running
Running
simonduerr
commited on
Commit
•
e37d702
1
Parent(s):
2818331
Update inference_app.py
Browse files- inference_app.py +139 -3
inference_app.py
CHANGED
@@ -5,16 +5,152 @@ import gradio as gr
|
|
5 |
|
6 |
from gradio_molecule3d import Molecule3D
|
7 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
|
|
|
|
|
|
|
9 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
|
11 |
def predict (input_sequence, input_ligand, input_protein):
|
12 |
start_time = time.time()
|
13 |
-
|
14 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
end_time = time.time()
|
16 |
run_time = end_time - start_time
|
17 |
-
return "
|
18 |
|
19 |
with gr.Blocks() as app:
|
20 |
|
|
|
5 |
|
6 |
from gradio_molecule3d import Molecule3D
|
7 |
|
8 |
+
import sys
|
9 |
+
import os
|
10 |
+
import os
|
11 |
+
import numpy as np
|
12 |
+
from rdkit import Chem
|
13 |
+
from rdkit.Chem import AllChem
|
14 |
+
from rdkit.Chem import Draw
|
15 |
+
from rdkit.Chem.Draw import IPythonConsole
|
16 |
+
from rdkit.Chem import DataStructs
|
17 |
+
from rdkit.Chem import RDConfig
|
18 |
+
from rdkit.Chem import rdBase
|
19 |
+
import pickle
|
20 |
|
21 |
+
from Bio.PDB import *
|
22 |
+
import requests
|
23 |
+
import subprocess
|
24 |
|
25 |
+
import mdtraj as md
|
26 |
+
from enspara import geometry
|
27 |
+
from sklearn.cluster import DBSCAN
|
28 |
+
import pandas as pd
|
29 |
+
|
30 |
+
def run_smina(
|
31 |
+
ligand_path, protein_path, out_path, pocket_center, pocket_size, num_poses=1, exhaustiveness=1
|
32 |
+
):
|
33 |
+
"""
|
34 |
+
Perform docking with Smina.
|
35 |
+
|
36 |
+
Parameters
|
37 |
+
----------
|
38 |
+
ligand_path: str or pathlib.Path
|
39 |
+
Path to ligand PDBQT file that should be docked.
|
40 |
+
protein_path: str or pathlib.Path
|
41 |
+
Path to protein PDBQT file that should be docked to.
|
42 |
+
out_path: str or pathlib.Path
|
43 |
+
Path to which docking poses should be saved, SDF or PDB format.
|
44 |
+
pocket_center: iterable of float or int
|
45 |
+
Coordinates defining the center of the binding site.
|
46 |
+
pocket_size: iterable of float or int
|
47 |
+
Lengths of edges defining the binding site.
|
48 |
+
num_poses: int
|
49 |
+
Maximum number of poses to generate.
|
50 |
+
exhaustiveness: int
|
51 |
+
Accuracy of docking calculations.
|
52 |
+
|
53 |
+
Returns
|
54 |
+
-------
|
55 |
+
output_text: str
|
56 |
+
The output of the Smina calculation.
|
57 |
+
"""
|
58 |
+
output_text = subprocess.check_output(
|
59 |
+
[
|
60 |
+
"./smina.static",
|
61 |
+
"--ligand",
|
62 |
+
str(ligand_path),
|
63 |
+
"--receptor",
|
64 |
+
str(protein_path),
|
65 |
+
"--out",
|
66 |
+
str(out_path),
|
67 |
+
"--center_x",
|
68 |
+
str(pocket_center[0]),
|
69 |
+
"--center_y",
|
70 |
+
str(pocket_center[1]),
|
71 |
+
"--center_z",
|
72 |
+
str(pocket_center[2]),
|
73 |
+
"--size_x",
|
74 |
+
str(pocket_size[0]),
|
75 |
+
"--size_y",
|
76 |
+
str(pocket_size[1]),
|
77 |
+
"--size_z",
|
78 |
+
str(pocket_size[2]),
|
79 |
+
"--num_modes",
|
80 |
+
str(num_poses),
|
81 |
+
"--exhaustiveness",
|
82 |
+
str(exhaustiveness),
|
83 |
+
],
|
84 |
+
universal_newlines=True, # needed to capture output text
|
85 |
+
)
|
86 |
+
return output_text
|
87 |
|
88 |
def predict (input_sequence, input_ligand, input_protein):
|
89 |
start_time = time.time()
|
90 |
+
|
91 |
+
m=Chem.MolFromSmiles(input_ligand)
|
92 |
+
|
93 |
+
m2=Chem.AddHs(m)
|
94 |
+
AllChem.EmbedMolecule(m2)
|
95 |
+
AllChem.MMFFOptimizeMolecule(m2)
|
96 |
+
|
97 |
+
Chem.SDWriter("ligand.sdf").write(m2)
|
98 |
+
|
99 |
+
os.system(f"obabel {input_protein.name} -xr -O receptor.pdbqt")
|
100 |
+
os.system("obabel -isdf ligand.sdf -O ligand.pdbqt")
|
101 |
+
#Find pocket
|
102 |
+
pdb = md.load('receptor.pdb')
|
103 |
+
# run ligsite
|
104 |
+
pockets_xyz = geometry.pockets.get_pocket_cells(struct=pdb)
|
105 |
+
eps_value = 0.15
|
106 |
+
min_samples_value = 5
|
107 |
+
dbscan = DBSCAN(eps=eps_value, min_samples=min_samples_value)
|
108 |
+
labels = dbscan.fit_predict(pockets_xyz)
|
109 |
+
|
110 |
+
# Find the unique clusters and their sizes
|
111 |
+
unique_labels, counts = np.unique(labels, return_counts=True)
|
112 |
+
|
113 |
+
# Exclude noise points
|
114 |
+
valid_clusters = unique_labels[unique_labels != -1]
|
115 |
+
valid_counts = counts[unique_labels != -1]
|
116 |
+
|
117 |
+
# Find the cluster with the most points (highest density)
|
118 |
+
densest_cluster_label = valid_clusters[np.argmax(valid_counts)]
|
119 |
+
densest_cluster_points = pockets_xyz[labels == densest_cluster_label]
|
120 |
+
|
121 |
+
pocket_center = np.mean(densest_cluster_points, axis=0)
|
122 |
+
|
123 |
+
|
124 |
+
import pandas as pd
|
125 |
+
|
126 |
+
top_df = pd.DataFrame()
|
127 |
+
top_df['serial'] = list(range(densest_cluster_points.shape[0]))
|
128 |
+
top_df['name'] = 'PK'
|
129 |
+
top_df['element'] = 'H'
|
130 |
+
top_df['resSeq'] = list(range(densest_cluster_points.shape[0]))
|
131 |
+
top_df['resName'] = 'PCK'
|
132 |
+
top_df['chainID'] = 0
|
133 |
+
|
134 |
+
pocket_top = md.Topology.from_dataframe(top_df, np.array([]))
|
135 |
+
pocket_trj = md.Trajectory(xyz=densest_cluster_points, topology=pocket_top)
|
136 |
+
pocket_trj.save('./pockets_dense.pdb')
|
137 |
+
|
138 |
+
parser = PDBParser()
|
139 |
+
struc = parser.get_structure("X", "pockets_dense.pdb")
|
140 |
+
coords = [x.coord for x in struc.get_atoms()]
|
141 |
+
pocket_center = np.mean(coords, axis=0)
|
142 |
+
output_text = run_smina(
|
143 |
+
"ligand.pdbqt",
|
144 |
+
"receptor.pdbqt",
|
145 |
+
"docking_pose.pdb",
|
146 |
+
pocket_center,
|
147 |
+
[10,10,10],
|
148 |
+
)
|
149 |
+
os.system("pdb_rplresname -UNL:LIG docked_pose.pdb")
|
150 |
+
os.system("pdb_merge receptor.pdb docked_pose.pdb > output.pdb")
|
151 |
end_time = time.time()
|
152 |
run_time = end_time - start_time
|
153 |
+
return "output.pdb", run_time
|
154 |
|
155 |
with gr.Blocks() as app:
|
156 |
|