Spaces:
Sleeping
Sleeping
Split attention.py into attention.py and models.py
Browse files- hexviz/app.py +2 -3
- hexviz/attention.py +2 -51
- hexviz/models.py +57 -0
- tests/test_attention.py +2 -26
- tests/test_models.py +28 -0
hexviz/app.py
CHANGED
@@ -3,7 +3,8 @@ import stmol
|
|
3 |
import streamlit as st
|
4 |
from stmol import showmol
|
5 |
|
6 |
-
from hexviz.attention import
|
|
|
7 |
|
8 |
st.title("Attention Visualization on proteins")
|
9 |
|
@@ -12,8 +13,6 @@ Visualize attention weights on protein structures for the protein language model
|
|
12 |
Pick a PDB ID, layer and head to visualize attention.
|
13 |
"""
|
14 |
|
15 |
-
|
16 |
-
# Define list of model types
|
17 |
models = [
|
18 |
# Model(name=ModelType.ProtGPT2, layers=36, heads=20),
|
19 |
Model(name=ModelType.TAPE_BERT, layers=12, heads=12),
|
|
|
3 |
import streamlit as st
|
4 |
from stmol import showmol
|
5 |
|
6 |
+
from hexviz.attention import get_attention_pairs
|
7 |
+
from hexviz.models import Model, ModelType
|
8 |
|
9 |
st.title("Attention Visualization on proteins")
|
10 |
|
|
|
13 |
Pick a PDB ID, layer and head to visualize attention.
|
14 |
"""
|
15 |
|
|
|
|
|
16 |
models = [
|
17 |
# Model(name=ModelType.ProtGPT2, layers=36, heads=20),
|
18 |
Model(name=ModelType.TAPE_BERT, layers=12, heads=12),
|
hexviz/attention.py
CHANGED
@@ -6,24 +6,10 @@ from urllib import request
|
|
6 |
import streamlit as st
|
7 |
import torch
|
8 |
from Bio.PDB import PDBParser, Polypeptide, Structure
|
9 |
-
from
|
10 |
-
|
11 |
-
T5Tokenizer)
|
12 |
|
13 |
|
14 |
-
class ModelType(str, Enum):
|
15 |
-
TAPE_BERT = "TAPE-BERT"
|
16 |
-
PROT_T5 = "prot_t5_xl_half_uniref50-enc"
|
17 |
-
ZymCTRL = "ZymCTRL"
|
18 |
-
ProtGPT2 = "ProtGPT2"
|
19 |
-
|
20 |
-
|
21 |
-
class Model:
|
22 |
-
def __init__(self, name, layers, heads):
|
23 |
-
self.name: ModelType = name
|
24 |
-
self.layers: int = layers
|
25 |
-
self.heads: int = heads
|
26 |
-
|
27 |
@st.cache
|
28 |
def get_structure(pdb_code: str) -> Structure:
|
29 |
"""
|
@@ -51,41 +37,6 @@ def get_sequences(structure: Structure) -> List[str]:
|
|
51 |
sequences.append("".join(list(residues_single_letter)))
|
52 |
return sequences
|
53 |
|
54 |
-
@st.cache
|
55 |
-
def get_protT5() -> Tuple[T5Tokenizer, T5EncoderModel]:
|
56 |
-
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
57 |
-
tokenizer = T5Tokenizer.from_pretrained(
|
58 |
-
"Rostlab/prot_t5_xl_half_uniref50-enc", do_lower_case=False
|
59 |
-
)
|
60 |
-
|
61 |
-
model = T5EncoderModel.from_pretrained("Rostlab/prot_t5_xl_half_uniref50-enc").to(
|
62 |
-
device
|
63 |
-
)
|
64 |
-
|
65 |
-
model.full() if device == "cpu" else model.half()
|
66 |
-
|
67 |
-
return tokenizer, model
|
68 |
-
|
69 |
-
@st.cache
|
70 |
-
def get_tape_bert() -> Tuple[TAPETokenizer, ProteinBertModel]:
|
71 |
-
tokenizer = TAPETokenizer()
|
72 |
-
model = ProteinBertModel.from_pretrained('bert-base', output_attentions=True)
|
73 |
-
return tokenizer, model
|
74 |
-
|
75 |
-
@st.cache
|
76 |
-
def get_zymctrl() -> Tuple[AutoTokenizer, GPT2LMHeadModel]:
|
77 |
-
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
78 |
-
tokenizer = AutoTokenizer.from_pretrained('nferruz/ZymCTRL')
|
79 |
-
model = GPT2LMHeadModel.from_pretrained('nferruz/ZymCTRL').to(device)
|
80 |
-
return tokenizer, model
|
81 |
-
|
82 |
-
@st.cache
|
83 |
-
def get_protgpt2() -> Tuple[AutoTokenizer, GPT2LMHeadModel]:
|
84 |
-
device = torch.device('cuda')
|
85 |
-
tokenizer = AutoTokenizer.from_pretrained('nferruz/ProtGPT2')
|
86 |
-
model = GPT2LMHeadModel.from_pretrained('nferruz/ProtGPT2').to(device)
|
87 |
-
return tokenizer, model
|
88 |
-
|
89 |
@st.cache
|
90 |
def get_attention(
|
91 |
sequence: str, model_type: ModelType = ModelType.TAPE_BERT
|
|
|
6 |
import streamlit as st
|
7 |
import torch
|
8 |
from Bio.PDB import PDBParser, Polypeptide, Structure
|
9 |
+
from models import (ModelType, get_protgpt2, get_protT5, get_tape_bert,
|
10 |
+
get_zymctrl)
|
|
|
11 |
|
12 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
@st.cache
|
14 |
def get_structure(pdb_code: str) -> Structure:
|
15 |
"""
|
|
|
37 |
sequences.append("".join(list(residues_single_letter)))
|
38 |
return sequences
|
39 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
@st.cache
|
41 |
def get_attention(
|
42 |
sequence: str, model_type: ModelType = ModelType.TAPE_BERT
|
hexviz/models.py
ADDED
@@ -0,0 +1,57 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from enum import Enum
|
2 |
+
from typing import Tuple
|
3 |
+
|
4 |
+
import streamlit as st
|
5 |
+
import torch
|
6 |
+
from tape import ProteinBertModel, TAPETokenizer
|
7 |
+
from transformers import (AutoTokenizer, GPT2LMHeadModel, T5EncoderModel,
|
8 |
+
T5Tokenizer)
|
9 |
+
|
10 |
+
|
11 |
+
class ModelType(str, Enum):
|
12 |
+
TAPE_BERT = "TAPE-BERT"
|
13 |
+
PROT_T5 = "prot_t5_xl_half_uniref50-enc"
|
14 |
+
ZymCTRL = "ZymCTRL"
|
15 |
+
ProtGPT2 = "ProtGPT2"
|
16 |
+
|
17 |
+
|
18 |
+
class Model:
|
19 |
+
def __init__(self, name, layers, heads):
|
20 |
+
self.name: ModelType = name
|
21 |
+
self.layers: int = layers
|
22 |
+
self.heads: int = heads
|
23 |
+
|
24 |
+
@st.cache
|
25 |
+
def get_protT5() -> Tuple[T5Tokenizer, T5EncoderModel]:
|
26 |
+
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
27 |
+
tokenizer = T5Tokenizer.from_pretrained(
|
28 |
+
"Rostlab/prot_t5_xl_half_uniref50-enc", do_lower_case=False
|
29 |
+
)
|
30 |
+
|
31 |
+
model = T5EncoderModel.from_pretrained("Rostlab/prot_t5_xl_half_uniref50-enc").to(
|
32 |
+
device
|
33 |
+
)
|
34 |
+
|
35 |
+
model.full() if device == "cpu" else model.half()
|
36 |
+
|
37 |
+
return tokenizer, model
|
38 |
+
|
39 |
+
@st.cache
|
40 |
+
def get_tape_bert() -> Tuple[TAPETokenizer, ProteinBertModel]:
|
41 |
+
tokenizer = TAPETokenizer()
|
42 |
+
model = ProteinBertModel.from_pretrained('bert-base', output_attentions=True)
|
43 |
+
return tokenizer, model
|
44 |
+
|
45 |
+
@st.cache
|
46 |
+
def get_zymctrl() -> Tuple[AutoTokenizer, GPT2LMHeadModel]:
|
47 |
+
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
48 |
+
tokenizer = AutoTokenizer.from_pretrained('nferruz/ZymCTRL')
|
49 |
+
model = GPT2LMHeadModel.from_pretrained('nferruz/ZymCTRL').to(device)
|
50 |
+
return tokenizer, model
|
51 |
+
|
52 |
+
@st.cache
|
53 |
+
def get_protgpt2() -> Tuple[AutoTokenizer, GPT2LMHeadModel]:
|
54 |
+
device = torch.device('cuda')
|
55 |
+
tokenizer = AutoTokenizer.from_pretrained('nferruz/ProtGPT2')
|
56 |
+
model = GPT2LMHeadModel.from_pretrained('nferruz/ProtGPT2').to(device)
|
57 |
+
return tokenizer, model
|
tests/test_attention.py
CHANGED
@@ -1,11 +1,8 @@
|
|
1 |
import torch
|
2 |
from Bio.PDB.Structure import Structure
|
3 |
-
from transformers import (GPT2LMHeadModel, GPT2TokenizerFast, T5EncoderModel,
|
4 |
-
T5Tokenizer)
|
5 |
|
6 |
-
from hexviz.attention import (ModelType, get_attention,
|
7 |
-
|
8 |
-
unidirectional_sum_filtered)
|
9 |
|
10 |
|
11 |
def test_get_structure():
|
@@ -27,27 +24,6 @@ def test_get_sequences():
|
|
27 |
A, B = sequences
|
28 |
assert A[:3] == ["M", "R", "I"]
|
29 |
|
30 |
-
def test_get_protT5():
|
31 |
-
result = get_protT5()
|
32 |
-
|
33 |
-
assert result is not None
|
34 |
-
assert isinstance(result, tuple)
|
35 |
-
|
36 |
-
tokenizer, model = result
|
37 |
-
|
38 |
-
assert isinstance(tokenizer, T5Tokenizer)
|
39 |
-
assert isinstance(model, T5EncoderModel)
|
40 |
-
|
41 |
-
def test_get_zymctrl():
|
42 |
-
result = get_zymctrl()
|
43 |
-
|
44 |
-
assert result is not None
|
45 |
-
assert isinstance(result, tuple)
|
46 |
-
|
47 |
-
tokenizer, model = result
|
48 |
-
|
49 |
-
assert isinstance(tokenizer, GPT2TokenizerFast)
|
50 |
-
assert isinstance(model, GPT2LMHeadModel)
|
51 |
|
52 |
def test_get_attention_zymctrl():
|
53 |
|
|
|
1 |
import torch
|
2 |
from Bio.PDB.Structure import Structure
|
|
|
|
|
3 |
|
4 |
+
from hexviz.attention import (ModelType, get_attention, get_sequences,
|
5 |
+
get_structure, unidirectional_sum_filtered)
|
|
|
6 |
|
7 |
|
8 |
def test_get_structure():
|
|
|
24 |
A, B = sequences
|
25 |
assert A[:3] == ["M", "R", "I"]
|
26 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
|
28 |
def test_get_attention_zymctrl():
|
29 |
|
tests/test_models.py
ADDED
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
from transformers import (GPT2LMHeadModel, GPT2TokenizerFast, T5EncoderModel,
|
3 |
+
T5Tokenizer)
|
4 |
+
|
5 |
+
from hexviz.models import get_protT5, get_zymctrl
|
6 |
+
|
7 |
+
|
8 |
+
def test_get_protT5():
|
9 |
+
result = get_protT5()
|
10 |
+
|
11 |
+
assert result is not None
|
12 |
+
assert isinstance(result, tuple)
|
13 |
+
|
14 |
+
tokenizer, model = result
|
15 |
+
|
16 |
+
assert isinstance(tokenizer, T5Tokenizer)
|
17 |
+
assert isinstance(model, T5EncoderModel)
|
18 |
+
|
19 |
+
def test_get_zymctrl():
|
20 |
+
result = get_zymctrl()
|
21 |
+
|
22 |
+
assert result is not None
|
23 |
+
assert isinstance(result, tuple)
|
24 |
+
|
25 |
+
tokenizer, model = result
|
26 |
+
|
27 |
+
assert isinstance(tokenizer, GPT2TokenizerFast)
|
28 |
+
assert isinstance(model, GPT2LMHeadModel)
|