File size: 3,550 Bytes
3b99abb | 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 | import torch
import json
import random
from ..data_interface import register_dataset
from transformers import EsmTokenizer
from ..lmdb_dataset import *
@register_dataset
class SaprotClassificationDataset(LMDBDataset):
def __init__(self,
tokenizer: str,
use_bias_feature: bool = False,
max_length: int = 1024,
preset_label: int = None,
mask_struc_ratio: float = None,
mask_seed: int = 20000812,
plddt_threshold: float = None,
**kwargs):
"""
Args:
tokenizer: Path to tokenizer
use_bias_feature: If True, structure information will be used
max_length: Max length of sequence
preset_label: If not None, all labels will be set to this value
mask_struc_ratio: Ratio of masked structure tokens, replace structure tokens with "#"
mask_seed: Seed for mask_struc_ratio
plddt_threshold: If not None, mask structure tokens with pLDDT < threshold
**kwargs:
"""
super().__init__(**kwargs)
self.tokenizer = EsmTokenizer.from_pretrained(tokenizer)
self.max_length = max_length
self.use_bias_feature = use_bias_feature
self.preset_label = preset_label
self.mask_struc_ratio = mask_struc_ratio
self.mask_seed = mask_seed
self.plddt_threshold = plddt_threshold
def __getitem__(self, index):
entry = json.loads(self._get(index))
seq = entry['seq']
# Mask structure tokens
if self.mask_struc_ratio is not None:
tokens = self.tokenizer.tokenize(seq)
mask_candi = [i for i, t in enumerate(tokens) if t[-1] != "#"]
# Randomly shuffle the mask candidates and set seed to ensure mask is consistent
random.seed(self.mask_seed)
random.shuffle(mask_candi)
# Mask first n structure tokens
mask_num = int(len(mask_candi) * self.mask_struc_ratio)
for i in range(mask_num):
idx = mask_candi[i]
tokens[idx] = tokens[idx][:-1] + "#"
seq = "".join(tokens)
# Mask structure tokens with pLDDT < threshold
if self.plddt_threshold is not None:
plddt = entry["plddt"]
tokens = self.tokenizer.tokenize(seq)
seq = ""
for token, score in zip(tokens, plddt):
if score < self.plddt_threshold:
seq += token[:-1] + "#"
else:
seq += token
tokens = self.tokenizer.tokenize(seq)[:self.max_length]
seq = " ".join(tokens)
if self.use_bias_feature:
coords = {k: v[:self.max_length] for k, v in entry['coords'].items()}
else:
coords = None
label = entry["label"] if self.preset_label is None else self.preset_label
return seq, label, coords
def __len__(self):
return int(self._get("length"))
def collate_fn(self, batch):
seqs, label_ids, coords = tuple(zip(*batch))
label_ids = torch.tensor(label_ids, dtype=torch.long)
labels = {"labels": label_ids}
encoder_info = self.tokenizer.batch_encode_plus(seqs, return_tensors='pt', padding=True)
inputs = {"inputs": encoder_info}
if self.use_bias_feature:
inputs["coords"] = coords
return inputs, labels |