DiffLinker / app.py
igashov
handle nan values if linker size is small
c1152c1
raw
history blame
No virus
10.5 kB
import argparse
import gradio as gr
import numpy as np
import os
import torch
import subprocess
import output
from rdkit import Chem
from src import const
from src.visualizer import save_xyz_file
from src.datasets import get_dataloader, collate_with_fragment_edges, parse_molecule
from src.lightning import DDPM
from src.linker_size_lightning import SizeClassifier
N_SAMPLES = 5
parser = argparse.ArgumentParser()
parser.add_argument('--ip', type=str, default=None)
args = parser.parse_args()
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
os.makedirs("results", exist_ok=True)
os.makedirs("models", exist_ok=True)
size_gnn_path = 'models/geom_size_gnn.ckpt'
if not os.path.exists(size_gnn_path):
print('Downloading SizeGNN model...')
link = 'https://zenodo.org/record/7121300/files/geom_size_gnn.ckpt?download=1'
subprocess.run(f'wget {link} -O {size_gnn_path}', shell=True)
size_nn = SizeClassifier.load_from_checkpoint('models/geom_size_gnn.ckpt', map_location=device).eval().to(device)
print('Loaded SizeGNN model')
diffusion_path = 'models/geom_difflinker.ckpt'
if not os.path.exists(diffusion_path):
print('Downloading Diffusion model...')
link = 'https://zenodo.org/record/7121300/files/geom_difflinker.ckpt?download=1'
subprocess.run(f'wget {link} -O {diffusion_path}', shell=True)
ddpm = DDPM.load_from_checkpoint('models/geom_difflinker.ckpt', map_location=device).eval().to(device)
print('Loaded diffusion model')
def read_molecule_content(path):
with open(path, "r") as f:
return "".join(f.readlines())
def read_molecule(path):
if path.endswith('.pdb'):
return Chem.MolFromPDBFile(path, sanitize=False, removeHs=True)
elif path.endswith('.mol'):
return Chem.MolFromMolFile(path, sanitize=False, removeHs=True)
elif path.endswith('.mol2'):
return Chem.MolFromMol2File(path, sanitize=False, removeHs=True)
elif path.endswith('.sdf'):
return Chem.SDMolSupplier(path, sanitize=False, removeHs=True)[0]
raise Exception('Unknown file extension')
def show_input(input_file):
if input_file is None:
return ['', gr.Radio.update(visible=False, value='Sample 1')]
if isinstance(input_file, str):
path = input_file
else:
path = input_file.name
extension = path.split('.')[-1]
if extension not in ['sdf', 'pdb', 'mol', 'mol2']:
msg = output.INVALID_FORMAT_MSG.format(extension=extension)
return [
output.IFRAME_TEMPLATE.format(html=msg),
gr.Radio.update(visible=False)
]
try:
molecule = read_molecule_content(path)
except Exception as e:
return [
f'Could not read the molecule: {e}',
gr.Radio.update(visible=False)
]
html = output.INITIAL_RENDERING_TEMPLATE.format(molecule=molecule, fmt=extension)
return [
output.IFRAME_TEMPLATE.format(html=html),
gr.Radio.update(visible=False)
]
def draw_sample(idx, out_files):
in_file = out_files[0]
in_sdf = in_file if isinstance(in_file, str) else in_file.name
out_file = out_files[idx + 1]
out_sdf = out_file if isinstance(out_file, str) else out_file.name
input_fragments_content = read_molecule_content(in_sdf)
generated_molecule_content = read_molecule_content(out_sdf)
html = output.SAMPLES_RENDERING_TEMPLATE.format(
fragments=input_fragments_content,
fragments_fmt='sdf',
molecule=generated_molecule_content,
molecule_fmt='sdf',
)
return output.IFRAME_TEMPLATE.format(html=html)
def generate(input_file, n_steps, n_atoms):
if input_file is None:
return ''
path = input_file.name
extension = path.split('.')[-1]
if extension not in ['sdf', 'pdb', 'mol', 'mol2']:
msg = output.INVALID_FORMAT_MSG.format(extension=extension)
return output.IFRAME_TEMPLATE.format(html=msg)
try:
molecule = read_molecule(path)
molecule = Chem.RemoveAllHs(molecule)
name = '.'.join(path.split('/')[-1].split('.')[:-1])
inp_sdf = f'results/input_{name}.sdf'
except Exception as e:
return f'Could not read the molecule: {e}'
if molecule.GetNumAtoms() > 50:
return f'Too large molecule: upper limit is 50 heavy atoms'
with Chem.SDWriter(inp_sdf) as w:
w.write(molecule)
positions, one_hot, charges = parse_molecule(molecule, is_geom=True)
anchors = np.zeros_like(charges)
fragment_mask = np.ones_like(charges)
linker_mask = np.zeros_like(charges)
print('Read and parsed molecule')
dataset = [{
'uuid': '0',
'name': '0',
'positions': torch.tensor(positions, dtype=const.TORCH_FLOAT, device=device),
'one_hot': torch.tensor(one_hot, dtype=const.TORCH_FLOAT, device=device),
'charges': torch.tensor(charges, dtype=const.TORCH_FLOAT, device=device),
'anchors': torch.tensor(anchors, dtype=const.TORCH_FLOAT, device=device),
'fragment_mask': torch.tensor(fragment_mask, dtype=const.TORCH_FLOAT, device=device),
'linker_mask': torch.tensor(linker_mask, dtype=const.TORCH_FLOAT, device=device),
'num_atoms': len(positions),
}] * N_SAMPLES
dataloader = get_dataloader(dataset, batch_size=N_SAMPLES, collate_fn=collate_with_fragment_edges)
print('Created dataloader')
ddpm.edm.T = n_steps
assert ddpm.center_of_mass == 'fragments'
if n_atoms == 0:
def sample_fn(_data):
out, _ = size_nn.forward(_data, return_loss=False)
probabilities = torch.softmax(out, dim=1)
distribution = torch.distributions.Categorical(probs=probabilities)
samples = distribution.sample()
sizes = []
for label in samples.detach().cpu().numpy():
sizes.append(size_nn.linker_id2size[label])
sizes = torch.tensor(sizes, device=samples.device, dtype=torch.long)
return sizes
else:
def sample_fn(_data):
return torch.ones(_data['positions'].shape[0], device=device, dtype=torch.long) * n_atoms
for data in dataloader:
chain, node_mask = ddpm.sample_chain(data, sample_fn=sample_fn, keep_frames=1)
print('Generated linker')
x = chain[0][:, :, :ddpm.n_dims]
h = chain[0][:, :, ddpm.n_dims:]
# Put the molecule back to the initial orientation
pos_masked = data['positions'] * data['fragment_mask']
N = data['fragment_mask'].sum(1, keepdims=True)
mean = torch.sum(pos_masked, dim=1, keepdim=True) / N
x = x + mean * node_mask
names = [f'output_{i+1}_{name}' for i in range(N_SAMPLES)]
save_xyz_file('results', h, x, node_mask, names=names, is_geom=True, suffix='')
print('Saved XYZ files')
break
out_files = []
for i in range(N_SAMPLES):
out_xyz = f'results/output_{i+1}_{name}_.xyz'
out_sdf = f'results/output_{i+1}_{name}_.sdf'
subprocess.run(f'obabel {out_xyz} -O {out_sdf}', shell=True)
out_files.append(out_sdf)
print('Converted to SDF')
return [
draw_sample(0, out_files),
[inp_sdf] + out_files,
gr.Radio.update(visible=True, value='Sample 1')
]
demo = gr.Blocks()
with demo:
gr.Markdown('# DiffLinker: Equivariant 3D-Conditional Diffusion Model for Molecular Linker Design')
gr.Markdown(
'Given a set of disconnected fragments in 3D, '
'DiffLinker places missing atoms in between and designs a molecule incorporating all the initial fragments. '
'Our method can link an arbitrary number of fragments, requires no information on the attachment atoms '
'and linker size, and can be conditioned on the protein pockets.'
)
gr.Markdown(
'[**[Paper]**](https://arxiv.org/abs/2210.05274) '
'[**[Code]**](https://github.com/igashov/DiffLinker)'
)
with gr.Box():
with gr.Row():
with gr.Column():
gr.Markdown('## Input Fragments')
gr.Markdown('Upload the file with 3D-coordinates of the input fragments in .pdb, .mol2 or .sdf format:')
input_file = gr.File(file_count='single', label='Input Fragments')
n_steps = gr.Slider(minimum=10, maximum=500, label="Number of Denoising Steps", step=10)
n_atoms = gr.Slider(
minimum=0, maximum=20,
label="Linker Size: DiffLinker will predict it if set to 0",
step=1
)
examples = gr.Dataset(
components=[gr.File(visible=False)],
samples=[['examples/example_1.sdf'], ['examples/example_2.sdf']],
type='index',
)
button = gr.Button('Generate Linker!')
gr.Markdown('')
gr.Markdown('## Output Files')
gr.Markdown('Download files with the generated molecules here:')
output_files = gr.File(file_count='multiple', label='Output Files', interactive=False)
with gr.Column():
gr.Markdown('## Visualization')
# gr.Markdown('Below you will see input and output molecules')
samples = gr.Radio(
choices=['Sample 1', 'Sample 2', 'Sample 3', 'Sample 4', 'Sample 5'],
value='Sample 1',
type='index',
show_label=False,
visible=False,
interactive=True,
)
visualization = gr.HTML()
input_file.change(
fn=show_input,
inputs=[input_file],
outputs=[visualization, samples],
)
input_file.clear(
fn=lambda: [None, '', gr.Radio.update(visible=False)],
inputs=[],
outputs=[input_file, visualization, samples],
)
examples.click(
fn=lambda idx: [f'examples/example_{idx+1}.sdf', 10, 0] + show_input(f'examples/example_{idx+1}.sdf'),
inputs=[examples],
outputs=[input_file, n_steps, n_atoms, visualization, samples]
)
button.click(
fn=generate,
inputs=[input_file, n_steps, n_atoms],
outputs=[visualization, output_files, samples],
)
samples.change(
fn=draw_sample,
inputs=[samples, output_files],
outputs=[visualization],
)
demo.launch(server_name=args.ip)