Spaces:
Runtime error
Runtime error
GlandVergil
commited on
Commit
•
2e2a6fa
1
Parent(s):
5edb0de
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,180 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import re
|
2 |
+
import os, time, pickle
|
3 |
+
import torch
|
4 |
+
from omegaconf import OmegaConf
|
5 |
+
import hydra
|
6 |
+
import logging
|
7 |
+
from rfdiffusion.util import writepdb_multi, writepdb
|
8 |
+
from rfdiffusion.inference import utils as iu
|
9 |
+
from hydra.core.hydra_config import HydraConfig
|
10 |
+
import numpy as np
|
11 |
+
import random
|
12 |
+
import glob
|
13 |
+
import gradio as gr
|
14 |
+
def greet(mtf):
|
15 |
+
return "Hello " + name + "!!"
|
16 |
+
def make_deterministic(seed=0):
|
17 |
+
torch.manual_seed(seed)
|
18 |
+
np.random.seed(seed)
|
19 |
+
random.seed(seed)
|
20 |
+
|
21 |
+
|
22 |
+
@hydra.main(version_base=None, config_path="../config/inference", config_name="base")
|
23 |
+
def main(conf: HydraConfig) -> None:
|
24 |
+
log = logging.getLogger(__name__)
|
25 |
+
if conf.inference.deterministic:
|
26 |
+
make_deterministic()
|
27 |
+
|
28 |
+
# Check for available GPU and print result of check
|
29 |
+
if torch.cuda.is_available():
|
30 |
+
device_name = torch.cuda.get_device_name(torch.cuda.current_device())
|
31 |
+
log.info(f"Found GPU with device_name {device_name}. Will run RFdiffusion on {device_name}")
|
32 |
+
else:
|
33 |
+
log.info("////////////////////////////////////////////////")
|
34 |
+
log.info("///// NO GPU DETECTED! Falling back to CPU /////")
|
35 |
+
log.info("////////////////////////////////////////////////")
|
36 |
+
|
37 |
+
# Initialize sampler and target/contig.
|
38 |
+
sampler = iu.sampler_selector(conf)
|
39 |
+
|
40 |
+
# Loop over number of designs to sample.
|
41 |
+
design_startnum = sampler.inf_conf.design_startnum
|
42 |
+
if sampler.inf_conf.design_startnum == -1:
|
43 |
+
existing = glob.glob(sampler.inf_conf.output_prefix + "*.pdb")
|
44 |
+
indices = [-1]
|
45 |
+
for e in existing:
|
46 |
+
print(e)
|
47 |
+
m = re.match(".*_(\d+)\.pdb$", e)
|
48 |
+
print(m)
|
49 |
+
if not m:
|
50 |
+
continue
|
51 |
+
m = m.groups()[0]
|
52 |
+
indices.append(int(m))
|
53 |
+
design_startnum = max(indices) + 1
|
54 |
+
|
55 |
+
for i_des in range(design_startnum, design_startnum + sampler.inf_conf.num_designs):
|
56 |
+
if conf.inference.deterministic:
|
57 |
+
make_deterministic(i_des)
|
58 |
+
|
59 |
+
start_time = time.time()
|
60 |
+
out_prefix = f"{sampler.inf_conf.output_prefix}_{i_des}"
|
61 |
+
log.info(f"Making design {out_prefix}")
|
62 |
+
if sampler.inf_conf.cautious and os.path.exists(out_prefix + ".pdb"):
|
63 |
+
log.info(
|
64 |
+
f"(cautious mode) Skipping this design because {out_prefix}.pdb already exists."
|
65 |
+
)
|
66 |
+
continue
|
67 |
+
|
68 |
+
x_init, seq_init = sampler.sample_init()
|
69 |
+
denoised_xyz_stack = []
|
70 |
+
px0_xyz_stack = []
|
71 |
+
seq_stack = []
|
72 |
+
plddt_stack = []
|
73 |
+
|
74 |
+
x_t = torch.clone(x_init)
|
75 |
+
seq_t = torch.clone(seq_init)
|
76 |
+
# Loop over number of reverse diffusion time steps.
|
77 |
+
for t in range(int(sampler.t_step_input), sampler.inf_conf.final_step - 1, -1):
|
78 |
+
px0, x_t, seq_t, plddt = sampler.sample_step(
|
79 |
+
t=t, x_t=x_t, seq_init=seq_t, final_step=sampler.inf_conf.final_step
|
80 |
+
)
|
81 |
+
px0_xyz_stack.append(px0)
|
82 |
+
denoised_xyz_stack.append(x_t)
|
83 |
+
seq_stack.append(seq_t)
|
84 |
+
plddt_stack.append(plddt[0]) # remove singleton leading dimension
|
85 |
+
|
86 |
+
# Flip order for better visualization in pymol
|
87 |
+
denoised_xyz_stack = torch.stack(denoised_xyz_stack)
|
88 |
+
denoised_xyz_stack = torch.flip(
|
89 |
+
denoised_xyz_stack,
|
90 |
+
[
|
91 |
+
0,
|
92 |
+
],
|
93 |
+
)
|
94 |
+
px0_xyz_stack = torch.stack(px0_xyz_stack)
|
95 |
+
px0_xyz_stack = torch.flip(
|
96 |
+
px0_xyz_stack,
|
97 |
+
[
|
98 |
+
0,
|
99 |
+
],
|
100 |
+
)
|
101 |
+
|
102 |
+
# For logging -- don't flip
|
103 |
+
plddt_stack = torch.stack(plddt_stack)
|
104 |
+
|
105 |
+
# Save outputs
|
106 |
+
os.makedirs(os.path.dirname(out_prefix), exist_ok=True)
|
107 |
+
final_seq = seq_stack[-1]
|
108 |
+
|
109 |
+
# Output glycines, except for motif region
|
110 |
+
final_seq = torch.where(
|
111 |
+
torch.argmax(seq_init, dim=-1) == 21, 7, torch.argmax(seq_init, dim=-1)
|
112 |
+
) # 7 is glycine
|
113 |
+
|
114 |
+
bfacts = torch.ones_like(final_seq.squeeze())
|
115 |
+
# make bfact=0 for diffused coordinates
|
116 |
+
bfacts[torch.where(torch.argmax(seq_init, dim=-1) == 21, True, False)] = 0
|
117 |
+
# pX0 last step
|
118 |
+
out = f"{out_prefix}.pdb"
|
119 |
+
|
120 |
+
# Now don't output sidechains
|
121 |
+
writepdb(
|
122 |
+
out,
|
123 |
+
denoised_xyz_stack[0, :, :4],
|
124 |
+
final_seq,
|
125 |
+
sampler.binderlen,
|
126 |
+
chain_idx=sampler.chain_idx,
|
127 |
+
bfacts=bfacts,
|
128 |
+
)
|
129 |
+
|
130 |
+
# run metadata
|
131 |
+
trb = dict(
|
132 |
+
config=OmegaConf.to_container(sampler._conf, resolve=True),
|
133 |
+
plddt=plddt_stack.cpu().numpy(),
|
134 |
+
device=torch.cuda.get_device_name(torch.cuda.current_device())
|
135 |
+
if torch.cuda.is_available()
|
136 |
+
else "CPU",
|
137 |
+
time=time.time() - start_time,
|
138 |
+
)
|
139 |
+
if hasattr(sampler, "contig_map"):
|
140 |
+
for key, value in sampler.contig_map.get_mappings().items():
|
141 |
+
trb[key] = value
|
142 |
+
with open(f"{out_prefix}.trb", "wb") as f_out:
|
143 |
+
pickle.dump(trb, f_out)
|
144 |
+
|
145 |
+
if sampler.inf_conf.write_trajectory:
|
146 |
+
# trajectory pdbs
|
147 |
+
traj_prefix = (
|
148 |
+
os.path.dirname(out_prefix) + "/traj/" + os.path.basename(out_prefix)
|
149 |
+
)
|
150 |
+
os.makedirs(os.path.dirname(traj_prefix), exist_ok=True)
|
151 |
+
|
152 |
+
out = f"{traj_prefix}_Xt-1_traj.pdb"
|
153 |
+
writepdb_multi(
|
154 |
+
out,
|
155 |
+
denoised_xyz_stack,
|
156 |
+
bfacts,
|
157 |
+
final_seq.squeeze(),
|
158 |
+
use_hydrogens=False,
|
159 |
+
backbone_only=False,
|
160 |
+
chain_ids=sampler.chain_idx,
|
161 |
+
)
|
162 |
+
|
163 |
+
out = f"{traj_prefix}_pX0_traj.pdb"
|
164 |
+
writepdb_multi(
|
165 |
+
out,
|
166 |
+
px0_xyz_stack,
|
167 |
+
bfacts,
|
168 |
+
final_seq.squeeze(),
|
169 |
+
use_hydrogens=False,
|
170 |
+
backbone_only=False,
|
171 |
+
chain_ids=sampler.chain_idx,
|
172 |
+
)
|
173 |
+
|
174 |
+
log.info(f"Finished design in {(time.time()-start_time)/60:.2f} minutes")
|
175 |
+
|
176 |
+
|
177 |
+
if __name__ == "__main__":
|
178 |
+
main()
|
179 |
+
iface = gr.Interface(fn=greet, inputs="text", outputs="text")
|
180 |
+
iface.launch()
|