Spaces:
Running
on
T4
Running
on
T4
File size: 61,151 Bytes
e1a6cd9 de38e57 e1a6cd9 de38e57 e1a6cd9 135c853 de38e57 e1a6cd9 85bd48b de38e57 e1a6cd9 de38e57 965a7b8 10f2049 965a7b8 85bd48b f969e9c 8397910 85bd48b de38e57 f969e9c 8397910 85bd48b 46db817 de38e57 85bd48b de38e57 85bd48b 907900a 85bd48b 8397910 85bd48b 3279dbd 7e2da24 b4ecf2a 8397910 85bd48b 8397910 85bd48b 8397910 85bd48b f969e9c 3055c36 85bd48b f969e9c 8397910 f969e9c 349d16a 8397910 f969e9c 8397910 3055c36 349d16a 3055c36 8397910 3279dbd 8397910 3055c36 f969e9c 85bd48b f969e9c cb07643 3055c36 fb52643 349d16a 85bd48b 349d16a 3279dbd 349d16a 85bd48b 349d16a 3279dbd 349d16a 3055c36 d3fb0e8 3055c36 3279dbd f969e9c e1a6cd9 8397910 0975cd6 135c853 965a7b8 8397910 0975cd6 8397910 7b9daf5 8397910 e4eb08d 8397910 e1a6cd9 8397910 e1a6cd9 f969e9c 965a7b8 7f2c740 965a7b8 2aa7536 965a7b8 7f2c740 2aa7536 f00454f 7f2c740 87d2088 7133751 965a7b8 7133751 965a7b8 3279dbd de38e57 3279dbd de38e57 3279dbd de38e57 3279dbd de38e57 3279dbd de38e57 3279dbd de38e57 3279dbd de38e57 80ea1c9 965a7b8 6dda5ff 80ea1c9 965a7b8 de38e57 541edde 3279dbd 965a7b8 3279dbd d13186d 965a7b8 3279dbd 965a7b8 3279dbd 965a7b8 3279dbd 965a7b8 3279dbd 965a7b8 3279dbd 965a7b8 3279dbd 8397910 98bc8ee 3279dbd 8397910 135c853 7f2c740 965a7b8 80ea1c9 8397910 87d2088 8397910 e4eb08d 8397910 e1a6cd9 f969e9c e1a6cd9 f969e9c e1a6cd9 d0c3de7 f969e9c 98bc8ee f969e9c e1a6cd9 f969e9c e1a6cd9 f969e9c e1a6cd9 965a7b8 d13186d 965a7b8 80ea1c9 3279dbd e1a6cd9 f969e9c e1a6cd9 f969e9c 3279dbd e1a6cd9 f969e9c e1a6cd9 8397910 e1a6cd9 8397910 f969e9c e1a6cd9 f969e9c e1a6cd9 f969e9c 8397910 349d16a e1a6cd9 f969e9c fde9ec3 f969e9c e1a6cd9 f969e9c e1a6cd9 f969e9c e1a6cd9 f969e9c e1a6cd9 f969e9c e1a6cd9 f969e9c e1a6cd9 f969e9c e1a6cd9 f969e9c e1a6cd9 f969e9c e1a6cd9 f969e9c 8397910 e1a6cd9 f969e9c e1a6cd9 f969e9c e1a6cd9 f969e9c 8397910 349d16a f969e9c 8397910 f969e9c 3279dbd d13186d 8397910 e1a6cd9 f969e9c e1a6cd9 f969e9c e1a6cd9 f969e9c 3279dbd f969e9c 85bd48b 8397910 349d16a 3279dbd f969e9c e1a6cd9 3279dbd f969e9c 135c853 3279dbd 349d16a 135c853 99db958 3279dbd 3055c36 349d16a 3279dbd f25fb09 3279dbd 349d16a 8397910 349d16a 8397910 349d16a 8397910 f969e9c 8397910 85bd48b f969e9c 8397910 f969e9c 349d16a 3279dbd 349d16a 8397910 965a7b8 8397910 3279dbd 3055c36 3279dbd f969e9c 3279dbd 3055c36 3279dbd 349d16a 3055c36 3279dbd 349d16a 3279dbd 3055c36 3279dbd 349d16a 3279dbd 85bd48b f969e9c 349d16a 21af4a5 349d16a f969e9c 349d16a 3279dbd 349d16a f969e9c 349d16a f969e9c 8397910 f969e9c 349d16a 8397910 965a7b8 8397910 349d16a f969e9c 349d16a 965a7b8 f969e9c 965a7b8 f969e9c 965a7b8 f969e9c 965a7b8 f969e9c 965a7b8 f969e9c 349d16a 3279dbd 349d16a 85bd48b 349d16a 85bd48b f969e9c 3279dbd f969e9c 349d16a 965a7b8 3279dbd 965a7b8 349d16a f969e9c 965a7b8 f969e9c 965a7b8 8397910 965a7b8 8397910 349d16a 8397910 965a7b8 8397910 f969e9c 5866cec f969e9c 349d16a f969e9c 3279dbd f969e9c 3279dbd f969e9c e1a6cd9 f3fe96d 99db958 01bcdce 99db958 01bcdce 99db958 fd7bfb8 99db958 f969e9c e1a6cd9 f969e9c e1a6cd9 8397910 f969e9c e1a6cd9 f969e9c e1a6cd9 f969e9c 349d16a f969e9c c10a0ce f969e9c 3279dbd 965a7b8 8397910 0975cd6 8397910 0975cd6 8397910 c10a0ce 8397910 f969e9c 98bc8ee 3279dbd 965a7b8 de38e57 3279dbd f969e9c e1a6cd9 f969e9c 3279dbd f969e9c c10a0ce 3279dbd c10a0ce 3279dbd c10a0ce 3279dbd c10a0ce 3279dbd f969e9c 965a7b8 e1a6cd9 f969e9c 8397910 92c1221 8397910 85bd48b 8397910 3279dbd 8397910 c10a0ce 8397910 349d16a 85bd48b 8397910 349d16a 3279dbd 8397910 3279dbd 8397910 965a7b8 349d16a f969e9c 8397910 98bc8ee 965a7b8 8397910 349d16a 3279dbd f969e9c 85bd48b 349d16a f969e9c 30a7365 f969e9c e1a6cd9 f969e9c 135c853 f969e9c 0f3c708 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 |
import json, time, os, sys, glob
import urllib
import shutil
import warnings
import copy
import random
import re
import os.path
import torch
import ray
import jax
import gradio as gr
import pandas as pd
import numpy as np
import plotly.express as px
import jax.numpy as jnp
import tensorflow as tf
import matplotlib.pyplot as plt
import colabfold as cf
import plotly.graph_objects as go
import torch.nn as nn
import torch.nn.functional as F
import tempfile
if "/home/user/app/af_backprop" not in sys.path:
sys.path.append("/home/user/app/af_backprop")
# local only
if "/home/duerr/phd/08_Code/ProteinMPNN/af_backprop" not in sys.path:
sys.path.append("/home/duerr/phd/08_Code/ProteinMPNN/af_backprop")
from torch import optim
from torch.utils.data import DataLoader
from torch.utils.data.dataset import random_split, Subset
from moleculekit.molecule import Molecule
from alphafold.common import protein
from alphafold.data import pipeline
from alphafold.model import data, config
from alphafold.model import model as afmodel
from alphafold.common import residue_constants
import moleculekit
print(moleculekit.__version__)
from utils import *
sys.path.append("/home/user/app/ProteinMPNN/vanilla_proteinmpnn")
sys.path.append("/home/duerr/phd/08_Code/ProteinMPNN/ProteinMPNN/vanilla_proteinmpnn")
# tf.config.set_visible_devices([], "GPU")
def chain_break(idx_res, Ls, length=200):
# Minkyung's code
# add big enough number to residue index to indicate chain breaks
L_prev = 0
for L_i in Ls[:-1]:
idx_res[L_prev + L_i :] += length
L_prev += L_i
return idx_res
def clear_mem():
backend = jax.lib.xla_bridge.get_backend()
for buf in backend.live_buffers():
buf.delete()
print("Is cuda available", torch.cuda.is_available())
# stream = os.popen("nvcc --version")
# output = stream.read()
# print(output)
def setup_af(seq, model_name="model_5_ptm"):
clear_mem()
# setup model
cfg = config.model_config("model_5_ptm")
cfg.model.num_recycle = 0
cfg.data.common.num_recycle = 0
cfg.data.eval.max_msa_clusters = 1
cfg.data.common.max_extra_msa = 1
cfg.data.eval.masked_msa_replace_fraction = 0
cfg.model.global_config.subbatch_size = None
if os.path.exists("/home/duerr"):
datadir = "/home/duerr/phd/08_Code/ProteinMPNN"
else:
datadir = "/home/user/app/"
model_params = data.get_model_haiku_params(model_name=model_name, data_dir=datadir)
model_runner = afmodel.RunModel(cfg, model_params, is_training=False)
Ls = [len(s) for s in seq.split("/")]
seq = re.sub("[^A-Z]", "", seq.upper())
length = len(seq)
feature_dict = {
**pipeline.make_sequence_features(
sequence=seq, description="none", num_res=length
),
**pipeline.make_msa_features(msas=[[seq]], deletion_matrices=[[[0] * length]]),
}
feature_dict["residue_index"] = chain_break(feature_dict["residue_index"], Ls)
inputs = model_runner.process_features(feature_dict, random_seed=0)
def runner(seq, opt):
# update sequence
inputs = opt["inputs"]
inputs.update(opt["prev"])
update_seq(seq, inputs)
update_aatype(inputs["target_feat"][..., 1:], inputs)
# mask prediction
mask = seq.sum(-1)
inputs["seq_mask"] = inputs["seq_mask"].at[:].set(mask)
inputs["msa_mask"] = inputs["msa_mask"].at[:].set(mask)
inputs["residue_index"] = jnp.where(mask == 1, inputs["residue_index"], 0)
# get prediction
key = jax.random.PRNGKey(0)
outputs = model_runner.apply(opt["params"], key, inputs)
prev = {
"init_msa_first_row": outputs["representations"]["msa_first_row"][None],
"init_pair": outputs["representations"]["pair"][None],
"init_pos": outputs["structure_module"]["final_atom_positions"][None],
}
aux = {
"final_atom_positions": outputs["structure_module"]["final_atom_positions"],
"final_atom_mask": outputs["structure_module"]["final_atom_mask"],
"plddt": get_plddt(outputs),
"pae": get_pae(outputs),
"inputs": inputs,
"prev": prev,
}
return aux
return jax.jit(runner), {"inputs": inputs, "params": model_params}
def make_tied_positions_for_homomers(pdb_dict_list):
my_dict = {}
for result in pdb_dict_list:
all_chain_list = sorted(
[item[-1:] for item in list(result) if item[:9] == "seq_chain"]
) # A, B, C, ...
tied_positions_list = []
chain_length = len(result[f"seq_chain_{all_chain_list[0]}"])
for i in range(1, chain_length + 1):
temp_dict = {}
for j, chain in enumerate(all_chain_list):
temp_dict[chain] = [i] # needs to be a list
tied_positions_list.append(temp_dict)
my_dict[result["name"]] = tied_positions_list
return my_dict
def align_structures(pdb1, pdb2, lenRes, index, random_dir):
"""Take two structure and superimpose pdb1 on pdb2"""
import Bio.PDB
import subprocess
pdb_parser = Bio.PDB.PDBParser(QUIET=True)
# Get the structures
ref_structure = pdb_parser.get_structure("ref", pdb1)
sample_structure = pdb_parser.get_structure("sample", pdb2)
aligner = Bio.PDB.CEAligner()
aligner.set_reference(ref_structure)
aligner.align(sample_structure)
io = Bio.PDB.PDBIO()
io.set_structure(ref_structure)
io.save(f"{random_dir}/outputs/reference.pdb")
io.set_structure(sample_structure)
io.save(f"{random_dir}/outputs/out_{index}_aligned.pdb")
# Doing this to get around biopython CEALIGN bug
# subprocess.call("pymol -c -Q -r cealign.pml", shell=True)
return aligner.rms, f"{random_dir}/outputs/reference.pdb", f"{random_dir}/outputs/out_{index}_aligned.pdb"
def save_pdb(outs, filename, LEN):
"""save pdb coordinates"""
p = {
"residue_index": outs["inputs"]["residue_index"][0][:LEN],
"aatype": outs["inputs"]["aatype"].argmax(-1)[0][:LEN],
"atom_positions": outs["final_atom_positions"][:LEN],
"atom_mask": outs["final_atom_mask"][:LEN],
}
b_factors = 100.0 * outs["plddt"][:LEN, None] * p["atom_mask"]
p = protein.Protein(**p, b_factors=b_factors)
pdb_lines = protein.to_pdb(p)
with open(filename, "w") as f:
f.write(pdb_lines)
@ray.remote(num_gpus=1, max_calls=1)
def run_alphafold(sequences, num_recycles, random_dir):
recycles = int(num_recycles)
RUNNER, OPT = setup_af(sequences[0])
plddts = []
paes = []
for i, sequence in enumerate(sequences):
SEQ = re.sub("[^A-Z]", "", sequence.upper())
MAX_LEN = len(SEQ)
LEN = len(SEQ)
x = np.array([residue_constants.restype_order.get(aa, -1) for aa in SEQ])
x = np.pad(x, [0, MAX_LEN - LEN], constant_values=-1)
x = jax.nn.one_hot(x, 20)
OPT["prev"] = {
"init_msa_first_row": np.zeros([1, MAX_LEN, 256]),
"init_pair": np.zeros([1, MAX_LEN, MAX_LEN, 128]),
"init_pos": np.zeros([1, MAX_LEN, 37, 3]),
}
positions = []
for r in range(recycles + 1):
outs = RUNNER(x, OPT)
outs = jax.tree_map(lambda x: np.asarray(x), outs)
positions.append(outs["prev"]["init_pos"][0, :LEN])
OPT["prev"] = outs["prev"]
plddts.append(outs["plddt"][:LEN])
paes.append(outs["pae"])
if os.path.exists("/home/duerr/phd/08_Code/ProteinMPNN"):
save_pdb(
outs, f"/home/duerr/phd/08_Code/ProteinMPNN/outputs/out_{i}.pdb", LEN
)
else:
print(f"saving to {random_dir.name}")
os.system(f"mkdir -p {random_dir.name}/outputs/")
save_pdb(outs, f"{random_dir.name}/outputs/out_{i}.pdb", LEN)
return plddts, paes, LEN
def setup_proteinmpnn(model_name="vanilla—v_48_020", backbone_noise=0.00):
from protein_mpnn_utils import (
loss_nll,
loss_smoothed,
gather_edges,
gather_nodes,
gather_nodes_t,
cat_neighbors_nodes,
_scores,
_S_to_seq,
tied_featurize,
parse_PDB,
)
from protein_mpnn_utils import StructureDataset, StructureDatasetPDB, ProteinMPNN
device = torch.device(
"cpu"
) # torch.device("cuda:0" if (torch.cuda.is_available()) else "cpu") #fix for memory issues
# ProteinMPNN model name: v_48_002, v_48_010, v_48_020, v_48_030, v_32_002, v_32_010; v_32_020, v_32_030; v_48_010=version with 48 edges 0.10A noise
# Standard deviation of Gaussian noise to add to backbone atoms
hidden_dim = 128
num_layers = 3
model, model_name = model_name.split("—")
if os.path.exists("/home/duerr"):
dir = "/home/duerr/phd/08_Code/ProteinMPNN"
else:
dir = "/home/user/app"
path_to_model_weights = (
f"{dir}/ProteinMPNN/{model}_model_weights"
)
model_folder_path = path_to_model_weights
if model_folder_path[-1] != "/":
model_folder_path = model_folder_path + "/"
checkpoint_path = model_folder_path + f"{model_name}.pt"
print("using ProteinMPNN weights from: ", checkpoint_path)
checkpoint = torch.load(checkpoint_path, map_location=device)
noise_level_print = checkpoint["noise_level"]
model = ProteinMPNN(
num_letters=21,
node_features=hidden_dim,
edge_features=hidden_dim,
hidden_dim=hidden_dim,
num_encoder_layers=num_layers,
num_decoder_layers=num_layers,
augment_eps=float(backbone_noise),
k_neighbors=checkpoint["num_edges"],
)
model.to(device)
model.load_state_dict(checkpoint["model_state_dict"])
model.eval()
return model, device
def get_pdb(pdb_code="", filepath=""):
if pdb_code is None or pdb_code == "":
try:
return filepath.name
except AttributeError as e:
return None
else:
os.system(f"wget -qnc https://files.rcsb.org/view/{pdb_code}.pdb")
return f"{pdb_code}.pdb"
def preprocess_mol(pdb_code="", filepath=""):
print(pdb_code)
if pdb_code is None or pdb_code == "":
try:
print(filepath.name)
mol = Molecule(filepath.name)
except AttributeError as e:
return None
else:
os.system(f"wget -qnc https://files.rcsb.org/view/{pdb_code}.pdb")
print(os.getcwd())
print(os.listdir())
print(os.system(f"head -n20 {pdb_code}.pdb"))
mol = Molecule(f"{pdb_code}.pdb")
print("print molecule loaded")
random_dir = tempfile.TemporaryDirectory()
mol.write(f"{random_dir.name}/original.pdb")
# clean messy files and only include protein itself
mol.filter("protein")
# renumber using moleculekit 0...len(protein)
df = mol.renumberResidues(returnMapping=True)
# add proteinMPNN index col which used 1..len(chain), 1...len(chain)
indexes = []
for chain, g in df.groupby("chain"):
j = 1
for i, row in g.iterrows():
indexes.append(j)
j += 1
df["proteinMPNN_index"] = indexes
mol.write(f"{random_dir.name}/original.pdb")
return f"{random_dir.name}/cleaned.pdb", df, f"{random_dir.name}/original.pdb"
def assign_sasa(mol):
from moleculekit.projections.metricsasa import MetricSasa
metr = MetricSasa(mode="residue", filtersel="protein")
sasaR = metr.project(mol)[0]
is_prot = mol.atomselect("protein")
resids = pd.DataFrame.from_dict({"resid": mol.resid, "is_prot": is_prot})
new_masses = []
i_without_non_prot = 0
for i, g in resids.groupby((resids["resid"].shift() != resids["resid"]).cumsum()):
if g["is_prot"].unique()[0] == True:
g["sasa"] = sasaR[i_without_non_prot]
i_without_non_prot += 1
else:
g["sasa"] = 0
new_masses.extend(list(g.sasa))
return np.array(new_masses)
def process_atomsel(atomsel):
"""everything lowercase and replace some keywords not relevant for protein design"""
atomsel = re.sub("sasa", "mass", atomsel, flags=re.I)
atomsel = re.sub("plddt", "beta", atomsel, flags=re.I)
return atomsel
def make_fixed_positions_dict(original_file, atomsel, residue_index_df):
# we use the uploaded file for the selection
print("fixed_pos using", original_file)
print(os.system(f"head -n10 {original_file}"))
mol = Molecule(original_file)
# use index for selection as resids will change
# set sasa to 0 for all non protein atoms (all non protein atoms are deleted later)
mol.masses = assign_sasa(mol)
print(mol.masses.shape)
print(assign_sasa(mol).shape)
atomsel = process_atomsel(atomsel)
selected_residues = mol.get("index", atomsel)
# clean up
mol.filter("protein")
mol.renumberResidues()
# based on selected index now get resids
selected_residues = [str(i) for i in selected_residues]
if len(selected_residues) == 0:
return None, []
selected_residues_str = " ".join(selected_residues)
selected_residues = set(mol.get("resid", sel=f"index {selected_residues_str}"))
# use the proteinMPNN index nomenclature to assemble fixed_positions_dict
fixed_positions_df = residue_index_df[
residue_index_df["new_resid"].isin(selected_residues)
]
chains = set(mol.get("chain", sel="all"))
fixed_position_dict = {"cleaned": {}}
# store the selected residues in a list for the visualization later with cleaned.pdb
selected_residues = list(fixed_positions_df["new_resid"])
for c in chains:
fixed_position_dict["cleaned"][c] = []
for i, row in fixed_positions_df.iterrows():
fixed_position_dict["cleaned"][row["chain"]].append(row["proteinMPNN_index"])
return fixed_position_dict, selected_residues
def update(
inp,
file,
designed_chain,
fixed_chain,
homomer,
num_seqs,
sampling_temp,
model_name,
backbone_noise,
omit_AAs,
atomsel,
):
from protein_mpnn_utils import (
loss_nll,
loss_smoothed,
gather_edges,
gather_nodes,
gather_nodes_t,
cat_neighbors_nodes,
_scores,
_S_to_seq,
tied_featurize,
parse_PDB,
)
from protein_mpnn_utils import StructureDataset, StructureDatasetPDB, ProteinMPNN
#pdb_path = get_pdb(pdb_code=inp, filepath=file)
pdb_path, mol_index, path_unprocessed = preprocess_mol(pdb_code=inp,filepath=file)
print("done processing mol")
if pdb_path == None:
return "Error processing PDB"
model, device = setup_proteinmpnn(
model_name=model_name, backbone_noise=float(backbone_noise)
)
if designed_chain == "":
designed_chain_list = []
else:
designed_chain_list = re.sub("[^A-Za-z]+", ",", designed_chain).split(",")
if fixed_chain == "":
fixed_chain_list = []
else:
fixed_chain_list = re.sub("[^A-Za-z]+", ",", fixed_chain).split(",")
chain_list = list(set(designed_chain_list + fixed_chain_list))
num_seq_per_target = int(num_seqs)
save_score = 0 # 0 for False, 1 for True; save score=-log_prob to npy files
save_probs = (
0 # 0 for False, 1 for True; save MPNN predicted probabilites per position
)
score_only = 0 # 0 for False, 1 for True; score input backbone-sequence pairs
conditional_probs_only = 0 # 0 for False, 1 for True; output conditional probabilities p(s_i given the rest of the sequence and backbone)
conditional_probs_only_backbone = 0 # 0 for False, 1 for True; if true output conditional probabilities p(s_i given backbone)
batch_size = 1 # Batch size; can set higher for titan, quadro GPUs, reduce this if running out of GPU memory
max_length = 20000 # Max sequence length
out_folder = "." # Path to a folder to output sequences, e.g. /home/out/
jsonl_path = "" # Path to a folder with parsed pdb into jsonl
if omit_AAs == "":
omit_AAs = "X" # Specify which amino acids should be omitted in the generated sequence, e.g. 'AC' would omit alanine and cystine.
pssm_multi = 0.0 # A value between [0.0, 1.0], 0.0 means do not use pssm, 1.0 ignore MPNN predictions
pssm_threshold = 0.0 # A value between -inf + inf to restric per position AAs
pssm_log_odds_flag = 0 # 0 for False, 1 for True
pssm_bias_flag = 0 # 0 for False, 1 for True
folder_for_outputs = out_folder
NUM_BATCHES = num_seq_per_target // batch_size
BATCH_COPIES = batch_size
temperatures = [sampling_temp]
omit_AAs_list = omit_AAs
alphabet = "ACDEFGHIKLMNPQRSTVWYX"
omit_AAs_np = np.array([AA in omit_AAs_list for AA in alphabet]).astype(np.float32)
chain_id_dict = None
if atomsel == "":
fixed_positions_dict, selected_residues = None, []
else:
fixed_positions_dict, selected_residues = make_fixed_positions_dict(path_unprocessed,
atomsel, mol_index
)
pssm_dict = None
omit_AA_dict = None
bias_AA_dict = None
bias_by_res_dict = None
bias_AAs_np = np.zeros(len(alphabet))
###############################################################
pdb_dict_list = parse_PDB(pdb_path, input_chain_list=chain_list)
dataset_valid = StructureDatasetPDB(
pdb_dict_list, truncate=None, max_length=max_length
)
if homomer:
tied_positions_dict = make_tied_positions_for_homomers(pdb_dict_list)
else:
tied_positions_dict = None
chain_id_dict = {}
chain_id_dict[pdb_dict_list[0]["name"]] = (designed_chain_list, fixed_chain_list)
with torch.no_grad():
for ix, prot in enumerate(dataset_valid):
score_list = []
all_probs_list = []
all_log_probs_list = []
S_sample_list = []
batch_clones = [copy.deepcopy(prot) for i in range(BATCH_COPIES)]
(
X,
S,
mask,
lengths,
chain_M,
chain_encoding_all,
chain_list_list,
visible_list_list,
masked_list_list,
masked_chain_length_list_list,
chain_M_pos,
omit_AA_mask,
residue_idx,
dihedral_mask,
tied_pos_list_of_lists_list,
pssm_coef,
pssm_bias,
pssm_log_odds_all,
bias_by_res_all,
tied_beta,
) = tied_featurize(
batch_clones,
device,
chain_id_dict,
fixed_positions_dict,
omit_AA_dict,
tied_positions_dict,
pssm_dict,
bias_by_res_dict,
)
pssm_log_odds_mask = (
pssm_log_odds_all > pssm_threshold
).float() # 1.0 for true, 0.0 for false
name_ = batch_clones[0]["name"]
randn_1 = torch.randn(chain_M.shape, device=X.device)
log_probs = model(
X,
S,
mask,
chain_M * chain_M_pos,
residue_idx,
chain_encoding_all,
randn_1,
)
mask_for_loss = mask * chain_M * chain_M_pos
scores = _scores(S, log_probs, mask_for_loss)
native_score = scores.cpu().data.numpy()
message = ""
seq_list = []
seq_recovery = []
seq_score = []
for temp in temperatures:
for j in range(NUM_BATCHES):
randn_2 = torch.randn(chain_M.shape, device=X.device)
if tied_positions_dict == None:
sample_dict = model.sample(
X,
randn_2,
S,
chain_M,
chain_encoding_all,
residue_idx,
mask=mask,
temperature=float(temp),
omit_AAs_np=omit_AAs_np,
bias_AAs_np=bias_AAs_np,
chain_M_pos=chain_M_pos,
omit_AA_mask=omit_AA_mask,
pssm_coef=pssm_coef,
pssm_bias=pssm_bias,
pssm_multi=pssm_multi,
pssm_log_odds_flag=bool(pssm_log_odds_flag),
pssm_log_odds_mask=pssm_log_odds_mask,
pssm_bias_flag=bool(pssm_bias_flag),
bias_by_res=bias_by_res_all,
)
S_sample = sample_dict["S"]
else:
sample_dict = model.tied_sample(
X,
randn_2,
S,
chain_M,
chain_encoding_all,
residue_idx,
mask=mask,
temperature=temp,
omit_AAs_np=omit_AAs_np,
bias_AAs_np=bias_AAs_np,
chain_M_pos=chain_M_pos,
omit_AA_mask=omit_AA_mask,
pssm_coef=pssm_coef,
pssm_bias=pssm_bias,
pssm_multi=pssm_multi,
pssm_log_odds_flag=bool(pssm_log_odds_flag),
pssm_log_odds_mask=pssm_log_odds_mask,
pssm_bias_flag=bool(pssm_bias_flag),
tied_pos=tied_pos_list_of_lists_list[0],
tied_beta=tied_beta,
bias_by_res=bias_by_res_all,
)
# Compute scores
S_sample = sample_dict["S"]
log_probs = model(
X,
S_sample,
mask,
chain_M * chain_M_pos,
residue_idx,
chain_encoding_all,
randn_2,
use_input_decoding_order=True,
decoding_order=sample_dict["decoding_order"],
)
mask_for_loss = mask * chain_M * chain_M_pos
scores = _scores(S_sample, log_probs, mask_for_loss)
scores = scores.cpu().data.numpy()
all_probs_list.append(sample_dict["probs"].cpu().data.numpy())
all_log_probs_list.append(log_probs.cpu().data.numpy())
S_sample_list.append(S_sample.cpu().data.numpy())
for b_ix in range(BATCH_COPIES):
masked_chain_length_list = masked_chain_length_list_list[b_ix]
masked_list = masked_list_list[b_ix]
seq_recovery_rate = torch.sum(
torch.sum(
torch.nn.functional.one_hot(S[b_ix], 21)
* torch.nn.functional.one_hot(S_sample[b_ix], 21),
axis=-1,
)
* mask_for_loss[b_ix]
) / torch.sum(mask_for_loss[b_ix])
seq = _S_to_seq(S_sample[b_ix], chain_M[b_ix])
score = scores[b_ix]
score_list.append(score)
native_seq = _S_to_seq(S[b_ix], chain_M[b_ix])
if b_ix == 0 and j == 0 and temp == temperatures[0]:
start = 0
end = 0
list_of_AAs = []
for mask_l in masked_chain_length_list:
end += mask_l
list_of_AAs.append(native_seq[start:end])
start = end
native_seq = "".join(
list(np.array(list_of_AAs)[np.argsort(masked_list)])
)
l0 = 0
for mc_length in list(
np.array(masked_chain_length_list)[
np.argsort(masked_list)
]
)[:-1]:
l0 += mc_length
native_seq = native_seq[:l0] + "/" + native_seq[l0:]
l0 += 1
sorted_masked_chain_letters = np.argsort(
masked_list_list[0]
)
print_masked_chains = [
masked_list_list[0][i]
for i in sorted_masked_chain_letters
]
sorted_visible_chain_letters = np.argsort(
visible_list_list[0]
)
print_visible_chains = [
visible_list_list[0][i]
for i in sorted_visible_chain_letters
]
native_score_print = np.format_float_positional(
np.float32(native_score.mean()),
unique=False,
precision=4,
)
line = ">{}, score={}, fixed_chains={}, designed_chains={}, model_name={}\n{}\n".format(
name_,
native_score_print,
print_visible_chains,
print_masked_chains,
model_name,
native_seq,
)
message += f"{line}\n"
start = 0
end = 0
list_of_AAs = []
for mask_l in masked_chain_length_list:
end += mask_l
list_of_AAs.append(seq[start:end])
start = end
seq = "".join(
list(np.array(list_of_AAs)[np.argsort(masked_list)])
)
# add non designed chains to predicted sequence
l0 = 0
for mc_length in list(
np.array(masked_chain_length_list)[np.argsort(masked_list)]
)[:-1]:
l0 += mc_length
seq = seq[:l0] + "/" + seq[l0:]
l0 += 1
score_print = np.format_float_positional(
np.float32(score), unique=False, precision=4
)
seq_rec_print = np.format_float_positional(
np.float32(seq_recovery_rate.detach().cpu().numpy()),
unique=False,
precision=4,
)
chain_s = ""
if len(visible_list_list[0]) > 0:
chain_M_bool = chain_M.bool()
not_designed = _S_to_seq(S[b_ix], ~chain_M_bool[b_ix])
labels = (
chain_encoding_all[b_ix][~chain_M_bool[b_ix]]
.detach()
.cpu()
.numpy()
)
for c in set(labels):
chain_s += "/"
nd_mask = labels == c
for i, x in enumerate(not_designed):
if nd_mask[i]:
chain_s += x
seq_recovery.append(seq_rec_print)
seq_score.append(score_print)
line = (
">T={}, sample={}, score={}, seq_recovery={}\n{}\n".format(
temp, b_ix, score_print, seq_rec_print, seq
)
)
seq_list.append(seq + chain_s)
message += f"{line}\n"
if fixed_positions_dict != None:
message += f"\nfixed positions:* {fixed_positions_dict['cleaned']} \n\n*uses CHAIN:[1..len(chain)] residue numbering"
# somehow sequences still contain X, remove again
for i, x in enumerate(seq_list):
for aa in omit_AAs:
seq_list[i] = x.replace(aa, "")
all_probs_concat = np.concatenate(all_probs_list)
all_log_probs_concat = np.concatenate(all_log_probs_list)
np.savetxt("all_probs_concat.csv", all_probs_concat.mean(0).T, delimiter=",")
np.savetxt(
"all_log_probs_concat.csv",
np.exp(all_log_probs_concat).mean(0).T,
delimiter=",",
)
S_sample_concat = np.concatenate(S_sample_list)
fig = px.imshow(
np.exp(all_log_probs_concat).mean(0).T,
labels=dict(x="positions", y="amino acids", color="probability"),
y=list(alphabet),
template="simple_white",
)
fig.update_xaxes(side="top")
fig_tadjusted = px.imshow(
all_probs_concat.mean(0).T,
labels=dict(x="positions", y="amino acids", color="probability"),
y=list(alphabet),
template="simple_white",
)
fig_tadjusted.update_xaxes(side="top")
seq_dict = {"seq_list": seq_list, "recovery": seq_recovery, "seq_score": seq_score}
return (
message,
fig,
fig_tadjusted,
gr.File.update(value="all_log_probs_concat.csv", visible=True),
gr.File.update(value="all_probs_concat.csv", visible=True),
pdb_path,
gr.Dropdown.update(choices=seq_list),
selected_residues,
seq_dict,
)
def update_AF(seq_dict, pdb, num_recycles, selectedResidues):
# # run alphafold using ray
# plddts, pae, num_res = run_alphafold(
# startsequence, num_recycles
# )
allSeqs = seq_dict["seq_list"]
lenSeqs = len(allSeqs)
if len(allSeqs[0]) > 700:
return (
"""
<div class="p-4 mb-4 text-sm text-yellow-700 bg-orange-50 rounded-lg" role="alert">
<span class="font-medium">Sorry!</span> Currently only small proteins can be run in the server in order to reduce wait time. Try a protein <700 aa. Bigger proteins you can run on <a href="https://github.com/sokrypton/colabfold">ColabFold</a>
</div>
""",
plt.figure(),
plt.figure(),
)
random_dir = tempfile.TemporaryDirectory()
plddts, paes, num_res = ray.get(run_alphafold.remote(allSeqs, num_recycles, random_dir ))
sequences = {}
for i in range(lenSeqs):
rms, input_pdb, aligned_pdb = align_structures(
pdb, f"{random_dir.name}/outputs/out_{i}.pdb", num_res, i, random_dir.name
)
sequences[i] = {
"Seq": i,
"RMSD": f"{rms:.2f}",
"Score": seq_dict["seq_score"][i],
"Recovery": seq_dict["recovery"][i],
"Mean pLDDT": f"{np.mean(plddts[i]):.4f}",
}
results = pd.DataFrame.from_dict(sequences, orient="index")
print(results)
plots = []
for index, plddts_val in enumerate(plddts):
# if recycle == 0 or recycle == len(plddts) - 1:
# visible = True
# else:
# visible = "legendonly"
visible = True
plots.append(
go.Scatter(
x=np.arange(len(plddts_val)),
y=plddts_val,
hovertemplate="<i>pLDDT</i>: %{y:.2f} <br><i>Residue index:</i> %{x}<br>Sequence "
+ str(index),
name=f"seq {index}",
visible=visible,
)
)
plotAF_plddt = go.Figure(data=plots)
plotAF_plddt.update_layout(
title="pLDDT",
xaxis_title="Residue index",
yaxis_title="pLDDT",
height=500,
template="simple_white",
legend=dict(yanchor="bottom", y=0.01, xanchor="left", x=0.99),
)
pae_plots = []
for i, pae in enumerate(paes):
plt.figure()
plt.title(f"Predicted Aligned Error sequence {i}")
Ln = pae.shape[0]
plt.imshow(pae, cmap="bwr", vmin=0, vmax=30, extent=(0, Ln, Ln, 0))
plt.colorbar()
plt.xlabel("Scored residue")
plt.ylabel("Aligned residue")
plt.savefig(f"outputs/pae_plot_{i}.png", dpi=300)
plt.close()
pae_plots.append(f"outputs/pae_plot_{i}.png")
# doesnt work (likely because too large)
# plotAF_pae = px.imshow(
# pae,
# labels=dict(x="Scored residue", y="Aligned residue", color=""),
# template="simple_white",
# y=np.arange(len(plddts_val)),
# )
# plotAF_pae.write_html("test.html")
# plotAF_pae.update_layout(title="Predicted Aligned Error", template="simple_white")
return (
molecule(
input_pdb,
aligned_pdb,
lenSeqs,
num_res,
selectedResidues,
allSeqs,
sequences,
random_dir.name
),
plotAF_plddt,
pae_plots,
results,
)
def read_mol(molpath):
with open(molpath, "r") as fp:
lines = fp.readlines()
mol = ""
for l in lines:
mol += l
return mol
def molecule(
input_pdb, aligned_pdb, lenSeqs, num_res, selectedResidues, allSeqs, sequences, random_dir
):
mol = read_mol(f"{random_dir}/outputs/reference.pdb")
options = ""
pred_mol = "["
seqdata = "{"
selected = "selected"
for i in range(lenSeqs):
seqdata += (
str(i)
+ ': { "score": '
+ sequences[i]["Score"]
+ ', "rmsd": '
+ sequences[i]["RMSD"]
+ ', "recovery": '
+ sequences[i]["Recovery"]
+ ', "plddt": '
+ sequences[i]["Mean pLDDT"]
+ ', "seq":"'
+ allSeqs[i]
+ '"}'
)
options += f'<option {selected} value="{i}">sequence {i} </option>' # RMSD {sequences[i]["RMSD"]}, score {sequences[i]["Score"]}, recovery {sequences[i]["Recovery"]} pLDDT {sequences[i]["Mean pLDDT"]}
p = f"{random_dir}/outputs/out_{i}_aligned.pdb"
pred_mol += f"`{read_mol(p)}`"
selected = ""
if i != lenSeqs - 1:
pred_mol += ","
seqdata += ","
pred_mol += "]"
seqdata += "}"
x = (
"""<!DOCTYPE html>
<html>
<head>
<meta http-equiv="content-type" content="text/html; charset=UTF-8" />
<link rel="stylesheet" href="https://unpkg.com/flowbite@1.4.5/dist/flowbite.min.css" />
<style>
body{
font-family:sans-serif
}
.mol-container {
width: 100%;
height: 700px;
position: relative;
}
.space-x-2 > * + *{
margin-left: 0.5rem;
}
.p-1{
padding:0.5rem;
}
.w-4{
width:1rem;
}
.h-4{
height:1rem;
}
.mt-4{
margin-top:1rem;
}
.mol-container select{
background-image:None;
}
</style>
<script src="https://cdnjs.cloudflare.com/ajax/libs/jquery/3.6.3/jquery.min.js" integrity="sha512-STof4xm1wgkfm7heWqFJVn58Hm3EtS31XFaagaa8VMReCXAkQnJZ+jEy8PCC/iT18dFy95WcExNHFTqLyp72eQ==" crossorigin="anonymous" referrerpolicy="no-referrer"></script>
<script src="https://3Dmol.csb.pitt.edu/build/3Dmol-min.js"></script>
</head>
<body>
<div class="max-w-2xl flex items-center space-x-2 py-3">
<label for="seq"
class=" text-right whitespace-nowrap block text-base font-medium text-gray-900 dark:text-gray-400">Select
a sequence</label>
<select id="seq"
class="bg-gray-50 border border-gray-300 text-gray-900 text-sm rounded-lg focus:ring-blue-500 focus:border-blue-500 block w-full p-2.5 dark:bg-gray-700 dark:border-gray-600 dark:placeholder-gray-400 dark:text-white dark:focus:ring-blue-500 dark:focus:border-blue-500">
"""
+ options
+ """
</select>
</div>
<div class="font-mono bg-gray-100 py-3 px-2 font-sm rounded">
<code>> seq <span id="id"></span>, score <span id="score"></span>, RMSD <span id="seqrmsd"></span>, Recovery
<span id="recovery"></span>, pLDDT <span id="plddt"></span></code><br>
<p id="seqText" class="max-w-4xl font-xs block" style="word-break: break-all;">
</p>
</div>
<div id="container" class="mol-container"></div>
<div class="flex items-center">
<div class="px-4 pt-2">
<label for="sidechain" class="relative inline-flex items-center mb-4 cursor-pointer ">
<input id="sidechain" type="checkbox" class="sr-only peer">
<div class="w-11 h-6 bg-gray-200 rounded-full peer peer-focus:ring-4 peer-focus:ring-blue-300 dark:peer-focus:ring-blue-800 dark:bg-gray-700 peer-checked:after:translate-x-full peer-checked:after:border-white after:absolute after:top-0.5 after:left-[2px] after:bg-white after:border-gray-300 after:border after:rounded-full after:h-5 after:w-5 after:transition-all dark:border-gray-600 peer-checked:bg-blue-600"></div>
<span class="ml-3 text-sm font-medium text-gray-900 dark:text-gray-300">Show side chains</span>
</label>
</div>
<div class="px-4 pt-2">
<label for="startstructure" class="relative inline-flex items-center mb-4 cursor-pointer ">
<input id="startstructure" type="checkbox" class="sr-only peer" checked>
<div class="w-11 h-6 bg-gray-200 rounded-full peer peer-focus:ring-4 peer-focus:ring-blue-300 dark:peer-focus:ring-blue-800 dark:bg-gray-700 peer-checked:after:translate-x-full peer-checked:after:border-white after:absolute after:top-0.5 after:left-[2px] after:bg-white after:border-gray-300 after:border after:rounded-full after:h-5 after:w-5 after:transition-all dark:border-gray-600 peer-checked:bg-blue-600"></div>
<span class="ml-3 text-sm font-medium text-gray-900 dark:text-gray-300">Show input structure</span>
</label>
</div>
<button type="button" class="text-gray-900 bg-white hover:bg-gray-100 border border-gray-200 focus:ring-4 focus:outline-none focus:ring-gray-100 font-medium rounded-lg text-sm px-5 py-2.5 text-center inline-flex items-center dark:focus:ring-gray-600 dark:bg-gray-800 dark:border-gray-700 dark:text-white dark:hover:bg-gray-700 mr-2 mb-2" id="download">
<svg class="w-6 h-6 mr-2 -ml-1" fill="none" stroke="currentColor" viewBox="0 0 24 24" xmlns="http://www.w3.org/2000/svg"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M4 16v1a3 3 0 003 3h10a3 3 0 003-3v-1m-4-4l-4 4m0 0l-4-4m4 4V4"></path></svg>
Download predicted structure
</button>
</div>
<div class="text-sm">
<div> RMSD AlphaFold vs. native: <span id="rmsd"></span> Å computed using CEAlign on the aligned fragment</div>
</div>
<div class="text-sm flex items-start">
<div class="w-1/2">
<div class="font-medium mt-4 flex items-center space-x-2"><b>AF2 model of redesigned sequence</b></div>
<div>AlphaFold model confidence:</div>
<div class="flex space-x-2 py-1"><span class="w-4 h-4" style="background-color: rgb(0, 83, 214);"> </span><span class="legendlabel">Very high
(pLDDT > 90)</span></div>
<div class="flex space-x-2 py-1"><span class="w-4 h-4" style="background-color: rgb(101, 203, 243);"> </span><span class="legendlabel">Confident
(90 > pLDDT > 70)</span></div>
<div class="flex space-x-2 py-1"><span class="w-4 h-4" style="background-color: rgb(255, 219, 19);"> </span><span class="legendlabel">Low (70 >
pLDDT > 50)</span></div>
<div class="flex space-x-2 py-1"><span class="w-4 h-4" style="background-color: rgb(255, 125, 69);"> </span><span class="legendlabel">Very low
(pLDDT < 50)</span></div>
<div class="row column legendDesc"> AlphaFold produces a per-residue confidence
score (pLDDT) between 0 and 100. Some regions below 50 pLDDT may be unstructured in isolation.
</div>
</div>
<div class="w-1/2">
<div class="font-medium mt-4 flex items-center space-x-2"><b>Input structure </b><span class="w-4 h-4 bg-gray-300 inline-flex" ></span></div>
<div class="flex space-x-2 py-1"><span class="w-4 h-4" style="background-color:hotpink" > </span><span class="legendlabel">Fixed positions</span></div>
</div>
</div>
<script>
function drawStructures(i, selectedResidues) {
$("#rmsd").text(seqs[i]["rmsd"])
$("#seqText").text(seqs[i]["seq"])
$("#seqrmsd").text(seqs[i]["rmsd"])
$("#id").text(i)
$("#score").text(seqs[i]["score"])
$("#recovery").text(seqs[i]["recovery"])
$("#plddt").text(seqs[i]["plddt"])
viewer = $3Dmol.createViewer(element, config);
viewer.addModel(data[i], "pdb");
viewer.addModel(pdb, "pdb");
viewer.getModel(1).setStyle({}, { cartoon: { colorscheme: { prop: "resi", map: colors } } })
viewer.getModel(0).setStyle({}, { cartoon: { colorfunc: colorAlpha } });
viewer.zoomTo();
viewer.render();
viewer.zoom(0.8, 2000);
viewer.getModel(0).setHoverable({}, true,
function (atom, viewer, event, container) {
if (!atom.label) {
atom.label = viewer.addLabel(atom.resn + atom.resi + " pLDDT=" + atom.b, { position: atom, backgroundColor: "mintcream", fontColor: "black" });
}
},
function (atom, viewer) {
if (atom.label) {
viewer.removeLabel(atom.label);
delete atom.label;
}
}
);
}
let viewer = null;
let voldata = null;
let element = null;
let config = null;
let currentIndex = 0;
let seqs = """
+ seqdata
+ """
let data = """
+ pred_mol
+ """
let pdb = `"""
+ mol
+ """`
var selectedResidues = """
+ f"{selectedResidues}"
+ """
//AlphaFold code from https://gist.github.com/piroyon/30d1c1099ad488a7952c3b21a5bebc96
let colorAlpha = function (atom) {
if (atom.b < 50) {
return "OrangeRed";
} else if (atom.b < 70) {
return "Gold";
} else if (atom.b < 90) {
return "MediumTurquoise";
} else {
return "Blue";
}
};
let colors = {}
for (let i=0; i<"""
+ str(num_res)
+ """;i++){
if (selectedResidues.includes(i)){
colors[i]="hotpink"
}else{
colors[i]="lightgray"
}}
let colorFixedSidechain = function(atom){
if (selectedResidues.includes(atom.resi)){
return "hotpink"
}else if (atom.elem == "O"){
return "red"
}else if (atom.elem == "N"){
return "blue"
}else if (atom.elem == "S"){
return "yellow"
}else{
return "lightgray"
}
}
$(document).ready(function () {
element = $("#container");
config = { backgroundColor: "white" };
//viewer.ui.initiateUI();
drawStructures(currentIndex, selectedResidues)
$("#sidechain").change(function () {
if (this.checked) {
BB = ["C", "O", "N"]
if ($("#startstructure").prop("checked")) {
viewer.getModel(0).setStyle( {"and": [{resn: ["GLY", "PRO"], invert: true},{atom: BB, invert: true},]},{stick: {colorscheme: "WhiteCarbon", radius: 0.3}, cartoon: { colorfunc: colorAlpha }});
viewer.getModel(1).setStyle( {"and": [{resn: ["GLY", "PRO"], invert: true},{atom: BB, invert: true},]},{stick: {colorfunc:colorFixedSidechain, radius: 0.3}, cartoon: {colorscheme:{prop:"resi",map:colors} }});
}else{
viewer.getModel(0).setStyle( {"and": [{resn: ["GLY", "PRO"], invert: true},{atom: BB, invert: true},]},{stick: {colorscheme: "WhiteCarbon", radius: 0.3}, cartoon: { colorfunc: colorAlpha }});
viewer.getModel(1).setStyle();
}
viewer.render()
} else {
if ($("#startstructure").prop("checked")) {
viewer.getModel(0).setStyle({cartoon: { colorfunc: colorAlpha }});
viewer.getModel(1).setStyle({cartoon: {colorscheme:{prop:"resi",map:colors} }});
}else{
viewer.getModel(0).setStyle({cartoon: { colorfunc: colorAlpha }});
viewer.getModel(1).setStyle();
}
viewer.render()
}
});
$("#seq").change(function () {
drawStructures(this.value, selectedResidues)
currentIndex = this.value
$("#sidechain").prop( "checked", false );
$("#startstructure").prop( "checked", true );
});
$("#startstructure").change(function () {
if (this.checked) {
$("#sidechain").prop( "checked", false );
viewer.getModel(1).setStyle({},{cartoon: {colorscheme:{prop:"resi",map:colors} } })
viewer.getModel(0).setStyle({}, { cartoon: { colorfunc: colorAlpha } });
viewer.render()
} else {
$("#sidechain").prop( "checked", false );
viewer.getModel(1).setStyle({},{})
viewer.getModel(0).setStyle({}, { cartoon: { colorfunc: colorAlpha } });
viewer.render()
}
});
$("#download").click(function () {
download("outputs/out_" + currentIndex + "_aligned.pdb", data[currentIndex]);
})
});
function download(filename, text) {
var element = document.createElement("a");
element.setAttribute("href", "data:text/plain;charset=utf-8," + encodeURIComponent(text));
element.setAttribute("download", filename);
element.style.display = "none";
document.body.appendChild(element);
element.click();
document.body.removeChild(element);
}
</script>
</body></html>"""
)
return f"""<iframe style="width: 800px; height: 1300px" name="result" allow="midi; geolocation; microphone; camera;
display-capture; encrypted-media;" sandbox="allow-modals allow-forms
allow-scripts allow-same-origin allow-popups
allow-top-navigation-by-user-activation allow-downloads" allowfullscreen=""
allowpaymentrequest="" frameborder="0" srcdoc='{x}'></iframe>"""
def set_examples(example):
(
label,
inp,
designed_chain,
fixed_chain,
homomer,
num_seqs,
sampling_temp,
atomsel,
) = example
return [
label,
inp,
designed_chain,
fixed_chain,
homomer,
gr.Slider.update(value=num_seqs),
gr.Radio.update(value=sampling_temp),
atomsel,
]
proteinMPNN = gr.Blocks()
with proteinMPNN:
# gr.Markdown("# MAINTENANC, CURRENTLY NOT WORKING")
# gr.HTML("<span style='font-size:3em;color:red'>⚠</span>")
gr.Markdown("# ProteinMPNN")
gr.Markdown(
"""This model takes as input a protein structure and based on its backbone predicts new sequences that will fold into that backbone.
Optionally, we can run AlphaFold2 on the predicted sequence to check whether the predicted sequences adopt the same backbone.
If you use this space please cite the ProteinMPNN paper
> J. Dauparas, I. Anishchenko, N. Bennett, H. Bai, R. J. Ragotte, L. F. Milles, B. I. M. Wicky, A. Courbet, R. J. de Haas, N. Bethel, P. J. Y. Leung, T. F. Huddy, S. Pellock, D. Tischer, F. Chan, B. Koepnick, H. Nguyen, A. Kang, B. Sankaran, A. K. Bera, N. P. King, D. Baker, Robust deep learning–based protein sequence design using ProteinMPNN. Science 378, 49–56 (2022).
and this webapp:
> Simon L. Dürr. (2023). ProteinMPNN Gradio Webapp (v0.3). Zenodo. https://doi.org/10.5281/zenodo.7630417
"""
)
gr.Markdown("![](https://simonduerr.eu/ProteinMPNN.png)")
with gr.Tabs():
with gr.TabItem("Input"):
inp = gr.Textbox(
placeholder="PDB Code or upload file below", label="Input structure"
)
file = gr.File(file_count="single")
with gr.TabItem("Settings"):
with gr.Row():
designed_chain = gr.Textbox(value="A", label="Designed chain")
fixed_chain = gr.Textbox(
placeholder="Use commas to fix multiple chains", label="Fixed chain"
)
with gr.Row():
num_seqs = gr.Slider(
minimum=1, maximum=15, value=1, step=1, label="Number of sequences"
)
sampling_temp = gr.Radio(
choices=["0.1", "0.15", "0.2", "0.25", "0.3"],
value="0.1",
label="Sampling temperature",
)
gr.Markdown(
""" Sampling temperature for amino acids, `T=0.0` means taking argmax, `T>>1.0` means sample randomly. Suggested values `0.1, 0.15, 0.2, 0.25, 0.3`. Higher values will lead to more diversity.
"""
)
with gr.Row():
model_name = gr.Dropdown(
choices=[
"vanilla—v_48_002",
"vanilla—v_48_010",
"vanilla—v_48_020",
"vanilla—v_48_030",
"soluble—v_48_010",
"soluble—v_48_020",
],
label="Model",
value="vanilla—v_48_020",
)
backbone_noise = gr.Dropdown(
choices=["0", "0.02", "0.10", "0.20", "0.30"], label="Backbone noise", value="0"
)
with gr.Row():
homomer = gr.Checkbox(value=False, label="Homomer?")
gr.Markdown(
"for correct symmetric tying lenghts of homomer chains should be the same"
)
with gr.Row():
omit_AAs = gr.Textbox(
placeholder="Specify omitted amino acids ", label="Omitted amino acids"
)
gr.Markdown("## Fixed positions")
gr.Markdown(
"""You can fix important positions in the protein. Resid should be specified with the same numbering as in the input pdb file. The fixed residues will be highlighted in the output.
The [VMD selection](http://www.ks.uiuc.edu/Research/vmd/vmd-1.9.2/ug/node89.html) synthax is used. You can also select based on ligands or chains in the input structure to specify interfaces to be fixed.
- <code>within 5 of resid 94</code> All residues that have >1 atom closer than 5 Å to any atom of residue 94
- <code>name CA and within 5 of resid 94</code> All residues that have CA atom closer than 5 Å to any atom of residue 94
- <code>resid 94 96 119</code> Residues 94, 94 and 119
- <code>within 5 of resname ZN</code> All residues with any atom <5 Å of zinc ion
- <code>chain A and within 5 of chain B </code> All residues of chain A that are part of the interface with chain B
- <code>protein and within 5 of nucleic </code> All residues that bind to DNA (if present in structure)
- <code>not (chain A and within 5 of chain B) </code> only modify residues that are in the interface with the fixed chain, not further away
- <code>chain A or (chain B and sasa < 20) </code> Keep chain A and all core residues fixeds
- <code>pLDDT >70 </code> Redesign all residues with low pLDDT
Note that <code>sasa</code> and <code>pLDDT</code> selectors modify default VMD behavior. SASA is calculated using moleculekit and written to the mass attribute. Selections based on mass do not work.
pLDDT is an alias for beta, it only works correctly with structures that contain the appropriate values in the beta column of the PDB file. """
)
atomsel = gr.Textbox(
placeholder="Specify atom selection ", label="Fixed positions"
)
btn = gr.Button("Run")
label = gr.Textbox(label="Label", visible=False)
examples = gr.Dataset(
components=[
label,
inp,
designed_chain,
fixed_chain,
homomer,
num_seqs,
sampling_temp,
atomsel,
],
samples=[
["Homomer design", "1O91", "A,B,C", "", True, "2", "0.1", ""],
["Monomer design", "6MRR", "A", "", False, "2", "0.1", ""],
["Redesign of Homomer to Heteromer", "3HTN", "A,B", "C", False, "2", "0.1", ""],
[
"Redesign of MID1 scaffold keeping binding site fixed",
"3V1C",
"A,B",
"",
False,
"2",
"0.1",
"within 5 of resname ZN",
],
[
"Redesign of DNA binding protein",
"3JRD",
"A,B",
"",
False,
"2",
"0.1",
"within 8 of nucleic",
],
[
"Surface Redesign of miniprotein",
"7JZM",
"A,B",
"",
False,
"2",
"0.1",
"chain B or (chain A and sasa < 20)",
],
],
)
gr.Markdown("# Output")
with gr.Tabs():
with gr.TabItem("Designed sequences"):
out = gr.Textbox(label="Status")
with gr.TabItem("Amino acid probabilities"):
plot = gr.Plot()
all_log_probs = gr.File(visible=False)
with gr.TabItem("T adjusted probabilities"):
gr.Markdown("Sampling temperature adjusted amino acid probabilties")
plot_tadjusted = gr.Plot()
all_probs = gr.File(visible=False)
with gr.TabItem("Structure validation w/ AF2"):
gr.HTML(
"""
<div class="flex items-center p-2 bg-gradient-to-r from-yellow-400 via-red-500 to-pink-500 rounded-lg shadow-sm">
<div>
<p class="text-base text-gray-700 dark:text-gray-200">
Results might differ from DeepMind's published results.
Predictions are made using <code>model_5_ptm</code> and without MSA based on the selected single sequence (<code>designed_chain</code> + <code>fixed_chain</code>).
</p>
</div>
</div>
"""
)
with gr.Row():
with gr.Row():
chosen_seq = gr.Dropdown(
choices=[],
label="Select a sequence for validation",
visible=False,
)
num_recycles = gr.Dropdown(
choices=["0", "1", "3", "5"], value="3", label="num Recycles"
)
btnAF = gr.Button("Run AlphaFold on all sequences")
with gr.Row():
mol = gr.HTML()
with gr.Column():
gr.Markdown("## Metrics")
p = {
0: {
"Seq": "NA",
"RMSD": "NA",
"Score": "NA",
"Recovery": "NA",
"Mean pLDDT": "NA",
}
}
placeholder = pd.DataFrame.from_dict(p, orient="index")
results = gr.Dataframe(
placeholder,
interactive=False,
row_count=(1, "dynamic"),
headers=["Seq", "RMSD", "Score", "Recovery", "Mean pLDDT"],
)
plotAF_plddt = gr.Plot(label="pLDDT")
# remove maxh80 class from css
plotAF_pae = gr.Gallery(label="PAE plots") # gr.Plot(label="PAE")
tempFile = gr.Variable()
selectedResidues = gr.Variable()
seq_dict = gr.Variable()
btn.click(
fn=update,
inputs=[
inp,
file,
designed_chain,
fixed_chain,
homomer,
num_seqs,
sampling_temp,
model_name,
backbone_noise,
omit_AAs,
atomsel,
],
outputs=[
out,
plot,
plot_tadjusted,
all_log_probs,
all_probs,
tempFile,
chosen_seq,
selectedResidues,
seq_dict,
],
)
btnAF.click(
fn=update_AF,
inputs=[seq_dict, tempFile, num_recycles, selectedResidues],
outputs=[mol, plotAF_plddt, plotAF_pae, results],
)
examples.click(fn=set_examples, inputs=examples, outputs=examples._components)
gr.Markdown(
"""Citation: **Robust deep learning based protein sequence design using ProteinMPNN** <br>
Justas Dauparas, Ivan Anishchenko, Nathaniel Bennett, Hua Bai, Robert J. Ragotte, Lukas F. Milles, Basile I. M. Wicky, Alexis Courbet, Robbert J. de Haas, Neville Bethel, Philip J. Y. Leung, Timothy F. Huddy, Sam Pellock, Doug Tischer, Frederick Chan, Brian Koepnick, Hannah Nguyen, Alex Kang, Banumathi Sankaran, Asim Bera, Neil P. King, David Baker <br>
bioRxiv 2022.06.03.494563; doi: [10.1101/2022.06.03.494563](https://doi.org/10.1101/2022.06.03.494563) <br><br> Server built by [@simonduerr](https://twitter.com/simonduerr) and hosted by Huggingface"""
)
ray.init(runtime_env={"working_dir": "./af_backprop"})
proteinMPNN.launch()
|