Spaces:
Runtime error
Runtime error
File size: 6,998 Bytes
8918ac7 |
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 |
import torch
import os
import gc
import argparse
import biotite.structure.io as bsio
import pandas as pd
from tqdm import tqdm
from Bio import SeqIO
from transformers import AutoTokenizer, EsmForProteinFolding
from transformers.models.esm.openfold_utils.protein import to_pdb, Protein as OFProtein
from transformers.models.esm.openfold_utils.feats import atom14_to_atom37
def read_fasta(file_path, key):
return str(getattr(SeqIO.read(file_path, 'fasta'), key))
def read_multi_fasta(file_path):
"""
params:
file_path: path to a fasta file
return:
a dictionary of sequences
"""
sequences = {}
current_sequence = ''
with open(file_path, 'r') as file:
for line in file:
line = line.strip()
if line.startswith('>'):
if current_sequence:
sequences[header] = current_sequence
current_sequence = ''
header = line
else:
current_sequence += line
if current_sequence:
sequences[header] = current_sequence
return sequences
def convert_outputs_to_pdb(outputs):
final_atom_positions = atom14_to_atom37(outputs["positions"][-1], outputs)
outputs = {k: v.to("cpu").numpy() for k, v in outputs.items()}
final_atom_positions = final_atom_positions.cpu().numpy()
final_atom_mask = outputs["atom37_atom_exists"]
pdbs = []
for i in range(outputs["aatype"].shape[0]):
aa = outputs["aatype"][i]
pred_pos = final_atom_positions[i]
mask = final_atom_mask[i]
resid = outputs["residue_index"][i] + 1
pred = OFProtein(
aatype=aa,
atom_positions=pred_pos,
atom_mask=mask,
residue_index=resid,
b_factors=outputs["plddt"][i],
chain_index=outputs["chain_index"][i] if "chain_index" in outputs else None,
)
pdbs.append(to_pdb(pred))
return pdbs
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument("--sequence", type=str, default=None)
parser.add_argument("--fasta_file", type=str, default=None)
parser.add_argument("--fasta_chunk_num", type=int, default=None)
parser.add_argument("--fasta_chunk_id", type=int, default=None)
parser.add_argument("--fasta_dir", type=str, default=None)
parser.add_argument("--out_dir", type=str)
parser.add_argument("--out_file", type=str, default="result.pdb")
parser.add_argument("--out_info_file", type=str, default=None)
parser.add_argument("--fold_chunk_size", type=int)
args = parser.parse_args()
# model = esm.pretrained.esmfold_v1()
# model = model.eval().cuda()
tokenizer = AutoTokenizer.from_pretrained("facebook/esmfold_v1")
model = EsmForProteinFolding.from_pretrained("facebook/esmfold_v1", low_cpu_mem_usage=True)
model = model.cuda()
# model.esm = model.esm.half()
torch.backends.cuda.matmul.allow_tf32 = True
# Optionally, uncomment to set a chunk size for axial attention. This can help reduce memory.
# Lower sizes will have lower memory requirements at the cost of increased speed.
if args.fold_chunk_size is not None:
model.trunk.set_chunk_size(args.fold_chunk_size)
if args.fasta_file is not None:
unfold_proteins = []
seq_dict = read_multi_fasta(args.fasta_file)
os.makedirs(args.out_dir, exist_ok=True)
names, sequences = list(seq_dict.keys()), list(seq_dict.values())
if args.fasta_chunk_num is not None:
chunk_size = len(names) // args.fasta_chunk_num + 1
start = args.fasta_chunk_id * chunk_size
end = min((args.fasta_chunk_id + 1) * chunk_size, len(names))
names, sequences = names[start:end], sequences[start:end]
out_info_dict = {"name": [], "plddt": []}
bar = tqdm(zip(names, sequences))
for name, sequence in bar:
bar.set_description(name)
name = name[1:].split(" ")[0]
out_file = os.path.join(args.out_dir, f"{name}.ef.pdb")
if os.path.exists(out_file):
out_info_dict["name"].append(name)
struct = bsio.load_structure(out_file, extra_fields=["b_factor"])
out_info_dict["plddt"].append(struct.b_factor.mean())
continue
# Multimer prediction can be done with chains separated by ':'
try:
tokenized_input = tokenizer([sequence], return_tensors="pt", add_special_tokens=False)['input_ids'].cuda()
with torch.no_grad():
output = model(tokenized_input)
except Exception as e:
print(e)
print(f"Failed to predict {name}")
unfold_proteins.append(name)
continue
gc.collect()
pdb = convert_outputs_to_pdb(output)
with open(out_file, "w") as f:
f.write("\n".join(pdb))
out_info_dict["name"].append(name)
struct = bsio.load_structure(out_file, extra_fields=["b_factor"])
out_info_dict["plddt"].append(struct.b_factor.mean())
if args.out_info_file is not None:
pd.DataFrame(out_info_dict).to_csv(args.out_info_file, index=False)
if args.fasta_dir is not None:
os.makedirs(args.out_dir, exist_ok=True)
proteins = sorted(os.listdir(args.fasta_dir))
bar = tqdm(proteins)
for p in bar:
name = p[:-6]
bar.set_description(name)
out_file = os.path.join(args.out_dir, f"{name}.ef.pdb")
if os.path.exists(out_file):
continue
bar.set_description(p)
sequence = read_fasta(os.path.join(args.fasta_dir, p), "seq")
tokenized_input = tokenizer([sequence], return_tensors="pt", add_special_tokens=False)['input_ids'].cuda()
# Multimer prediction can be done with chains separated by ':'
with torch.no_grad():
output = model(tokenized_input)
pdb = convert_outputs_to_pdb(output)
with open(out_file, "w") as f:
f.write("\n".join(pdb))
struct = bsio.load_structure(out_file, extra_fields=["b_factor"])
print(p, struct.b_factor.mean())
elif args.sequence is not None:
sequence = args.sequence
# Multimer prediction can be done with chains separated by ':'
with torch.no_grad():
output = model.infer_pdb(sequence)
with open(args.out_file, "w") as f:
f.write(output)
struct = bsio.load_structure(args.out_file, extra_fields=["b_factor"])
print(struct.b_factor.mean()) |