Spaces:
Runtime error
Runtime error
xuyingli
commited on
Commit
·
481cfc6
1
Parent(s):
0617c9a
app.py
Browse filesAdd application file
app.py
ADDED
@@ -0,0 +1,481 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
import torch
|
3 |
+
import esm
|
4 |
+
import matplotlib.pyplot as plt
|
5 |
+
from myscaledb import Client
|
6 |
+
import random
|
7 |
+
from collections import Counter
|
8 |
+
from tqdm import tqdm
|
9 |
+
from statistics import mean
|
10 |
+
|
11 |
+
import torch
|
12 |
+
import matplotlib.pyplot as plt
|
13 |
+
import numpy as np
|
14 |
+
import pandas as pd
|
15 |
+
import seaborn as sns
|
16 |
+
from stmol import *
|
17 |
+
import py3Dmol
|
18 |
+
# from streamlit_3Dmol import component_3dmol
|
19 |
+
|
20 |
+
import esm
|
21 |
+
|
22 |
+
import scipy
|
23 |
+
from sklearn.model_selection import GridSearchCV, train_test_split
|
24 |
+
from sklearn.decomposition import PCA
|
25 |
+
from sklearn.neighbors import KNeighborsClassifier, KNeighborsRegressor
|
26 |
+
from sklearn.svm import SVC, SVR
|
27 |
+
from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor
|
28 |
+
from sklearn.naive_bayes import GaussianNB
|
29 |
+
from sklearn.linear_model import LogisticRegression, SGDRegressor
|
30 |
+
from sklearn.pipeline import Pipeline
|
31 |
+
|
32 |
+
from streamlit.components.v1 import html
|
33 |
+
|
34 |
+
|
35 |
+
def init_esm():
|
36 |
+
msa_transformer, msa_transformer_alphabet = esm.pretrained.esm_msa1b_t12_100M_UR50S()
|
37 |
+
msa_transformer = msa_transformer.eval()
|
38 |
+
return msa_transformer, msa_transformer_alphabet
|
39 |
+
|
40 |
+
@st.experimental_singleton(show_spinner=False)
|
41 |
+
def init_db():
|
42 |
+
""" Initialize the Database Connection
|
43 |
+
|
44 |
+
Returns:
|
45 |
+
meta_field: Meta field that records if an image is viewed
|
46 |
+
client: Database connection object
|
47 |
+
"""
|
48 |
+
client = Client(
|
49 |
+
url=st.secrets["DB_URL"], user=st.secrets["USER"], password=st.secrets["PASSWD"])
|
50 |
+
# We can check if the connection is alive
|
51 |
+
assert client.is_alive()
|
52 |
+
meta_field = {}
|
53 |
+
return meta_field, Client
|
54 |
+
|
55 |
+
|
56 |
+
def perdict_contact_visualization(seq, model, batch_converter):
|
57 |
+
data = [
|
58 |
+
("protein1", seq),
|
59 |
+
]
|
60 |
+
batch_labels, batch_strs, batch_tokens = batch_converter(data)
|
61 |
+
|
62 |
+
# Extract per-residue representations (on CPU)
|
63 |
+
with torch.no_grad():
|
64 |
+
results = model(batch_tokens, repr_layers=[12], return_contacts=True)
|
65 |
+
token_representations = results["representations"][12]
|
66 |
+
|
67 |
+
# Generate per-sequence representations via averaging
|
68 |
+
# NOTE: token 0 is always a beginning-of-sequence token, so the first residue is token 1.
|
69 |
+
|
70 |
+
sequence_representations = []
|
71 |
+
for i, (_, seq) in enumerate(data):
|
72 |
+
sequence_representations.append(token_representations[i, 1 : len(seq) + 1].mean(0))
|
73 |
+
|
74 |
+
# Look at the unsupervised self-attention map contact predictions
|
75 |
+
for (_, seq), attention_contacts in zip(data, results["contacts"]):
|
76 |
+
fig, ax = plt.subplots()
|
77 |
+
ax.matshow(attention_contacts[: len(seq), : len(seq)])
|
78 |
+
|
79 |
+
fig.suptitle(seq)
|
80 |
+
# fig.set_facecolor('black')
|
81 |
+
|
82 |
+
return fig
|
83 |
+
|
84 |
+
|
85 |
+
def visualize_3D_Coordinates(coords):
|
86 |
+
xs = []
|
87 |
+
ys = []
|
88 |
+
zs = []
|
89 |
+
for i in coords:
|
90 |
+
xs.append(i[0])
|
91 |
+
ys.append(i[1])
|
92 |
+
zs.append(i[2])
|
93 |
+
fig = plt.figure(figsize=(10,10))
|
94 |
+
ax = fig.add_subplot(111, projection='3d')
|
95 |
+
ax.set_title('3D coordinates of $C_{b}$ backbone structure')
|
96 |
+
N = len(coords)
|
97 |
+
for i in range(len(coords) - 1):
|
98 |
+
ax.plot(
|
99 |
+
xs[i:i+2], ys[i:i+2], zs[i:i+2],
|
100 |
+
color=plt.cm.viridis(i/N),
|
101 |
+
marker='o'
|
102 |
+
)
|
103 |
+
return fig
|
104 |
+
|
105 |
+
def esm_search(model, sequnce, batch_converter,top_k=5):
|
106 |
+
data = [
|
107 |
+
("protein1", sequnce),
|
108 |
+
]
|
109 |
+
batch_labels, batch_strs, batch_tokens = batch_converter(data)
|
110 |
+
|
111 |
+
# Extract per-residue representations (on CPU)
|
112 |
+
with torch.no_grad():
|
113 |
+
results = model(batch_tokens, repr_layers=[12], return_contacts=True)
|
114 |
+
token_representations = results["representations"][12]
|
115 |
+
|
116 |
+
token_list = token_representations.tolist()[0][0][0]
|
117 |
+
|
118 |
+
client = Client(
|
119 |
+
url=st.secrets["DB_URL"], user=st.secrets["USER"], password=st.secrets["PASSWD"])
|
120 |
+
|
121 |
+
result = client.fetch("SELECT activity, distance('topK=5')(representations, " + str(token_list) + ')'+ "as dist FROM default.esm_protein_indexer_768")
|
122 |
+
result_temp_seq = []
|
123 |
+
for i in result:
|
124 |
+
# print(result_temp_seq)
|
125 |
+
result_temp_coords = i['coords']
|
126 |
+
result_temp_seq.append(i['seq'])
|
127 |
+
|
128 |
+
return result_temp_coords, result_temp_seq
|
129 |
+
|
130 |
+
def KNN_search(sequence):
|
131 |
+
model, alphabet = esm.pretrained.esm2_t33_650M_UR50D()
|
132 |
+
batch_converter = alphabet.get_batch_converter()
|
133 |
+
model.eval()
|
134 |
+
data = [("protein1", sequence),
|
135 |
+
]
|
136 |
+
batch_labels, batch_strs, batch_tokens = batch_converter(data)
|
137 |
+
batch_lens = (batch_tokens != alphabet.padding_idx).sum(1)
|
138 |
+
with torch.no_grad():
|
139 |
+
results = model(batch_tokens, repr_layers=[33], return_contacts=True)
|
140 |
+
token_representations = results["representations"][33]
|
141 |
+
token_list = token_representations.tolist()[0][0]
|
142 |
+
print(token_list)
|
143 |
+
client = Client(
|
144 |
+
url=st.secrets["DB_URL"], user=st.secrets["USER"], password=st.secrets["PASSWD"])
|
145 |
+
|
146 |
+
result = client.fetch("SELECT activity, distance('topK=10')(representations, " + str(token_list) + ')'+ "as dist FROM default.esm_protein_indexer")
|
147 |
+
result_temp_activity = []
|
148 |
+
for i in result:
|
149 |
+
# print(result_temp_seq)
|
150 |
+
result_temp_activity.append(i['activity'])
|
151 |
+
|
152 |
+
res_1 = sum(result_temp_activity)/len(result_temp_activity)
|
153 |
+
return res_1
|
154 |
+
|
155 |
+
|
156 |
+
|
157 |
+
def train_test_split_PCA(dataset):
|
158 |
+
ys = []
|
159 |
+
Xs = []
|
160 |
+
FASTA_PATH = '/root/xuying_experiments/esm-main/P62593.fasta'
|
161 |
+
EMB_PATH = '/root/xuying_experiments/esm-main/P62593_reprs'
|
162 |
+
for header, _seq in esm.data.read_fasta(FASTA_PATH):
|
163 |
+
scaled_effect = header.split('|')[-1]
|
164 |
+
ys.append(float(scaled_effect))
|
165 |
+
fn = f'{EMB_PATH}/{header}.pt'
|
166 |
+
embs = torch.load(fn)
|
167 |
+
Xs.append(embs['mean_representations'][34])
|
168 |
+
|
169 |
+
Xs = torch.stack(Xs, dim=0).numpy()
|
170 |
+
train_size = 0.8
|
171 |
+
Xs_train, Xs_test, ys_train, ys_test = train_test_split(Xs, ys, train_size=train_size, random_state=42)
|
172 |
+
return Xs_train, Xs_test, ys_train, ys_test
|
173 |
+
|
174 |
+
def PCA_visual(Xs_train):
|
175 |
+
num_pca_components = 60
|
176 |
+
pca = PCA(num_pca_components)
|
177 |
+
Xs_train_pca = pca.fit_transform(Xs_train)
|
178 |
+
fig_dims = (4, 4)
|
179 |
+
fig, ax = plt.subplots(figsize=fig_dims)
|
180 |
+
ax.set_title('Visualize Embeddings')
|
181 |
+
sc = ax.scatter(Xs_train_pca[:,0], Xs_train_pca[:,1], c=ys_train, marker='.')
|
182 |
+
ax.set_xlabel('PCA first principal component')
|
183 |
+
ax.set_ylabel('PCA second principal component')
|
184 |
+
plt.colorbar(sc, label='Variant Effect')
|
185 |
+
|
186 |
+
return fig
|
187 |
+
|
188 |
+
def KNN_trainings(Xs_train, Xs_test, ys_train, ys_test):
|
189 |
+
num_pca_components = 60
|
190 |
+
knn_grid = [
|
191 |
+
{
|
192 |
+
'model': [KNeighborsRegressor()],
|
193 |
+
'model__n_neighbors': [5, 10],
|
194 |
+
'model__weights': ['uniform', 'distance'],
|
195 |
+
'model__algorithm': ['ball_tree', 'kd_tree', 'brute'],
|
196 |
+
'model__leaf_size' : [15, 30],
|
197 |
+
'model__p' : [1, 2],
|
198 |
+
}]
|
199 |
+
|
200 |
+
cls_list = [KNeighborsRegressor]
|
201 |
+
param_grid_list = [knn_grid]
|
202 |
+
|
203 |
+
pipe = Pipeline(
|
204 |
+
steps = (
|
205 |
+
('pca', PCA(num_pca_components)),
|
206 |
+
('model', KNeighborsRegressor())
|
207 |
+
)
|
208 |
+
)
|
209 |
+
|
210 |
+
result_list = []
|
211 |
+
grid_list = []
|
212 |
+
|
213 |
+
for cls_name, param_grid in zip(cls_list, param_grid_list):
|
214 |
+
print(cls_name)
|
215 |
+
grid = GridSearchCV(
|
216 |
+
estimator = pipe,
|
217 |
+
param_grid = param_grid,
|
218 |
+
scoring = 'r2',
|
219 |
+
verbose = 1,
|
220 |
+
n_jobs = -1 # use all available cores
|
221 |
+
)
|
222 |
+
grid.fit(Xs_train, ys_train)
|
223 |
+
# print(Xs_train, ys_train)
|
224 |
+
result_list.append(pd.DataFrame.from_dict(grid.cv_results_))
|
225 |
+
grid_list.append(grid)
|
226 |
+
|
227 |
+
dataframe = pd.DataFrame(result_list[0].sort_values('rank_test_score')[:5])
|
228 |
+
|
229 |
+
|
230 |
+
return dataframe[['param_model','params','param_model__algorithm','mean_test_score','rank_test_score']]
|
231 |
+
|
232 |
+
|
233 |
+
st.markdown("""
|
234 |
+
<link
|
235 |
+
rel="stylesheet"
|
236 |
+
href="https://fonts.googleapis.com/css?family=Roboto:300,400,500,700&display=swap"
|
237 |
+
/>
|
238 |
+
""", unsafe_allow_html=True)
|
239 |
+
|
240 |
+
messages = [
|
241 |
+
f"""
|
242 |
+
Evolutionary-scale prediction of atomic level protein structure
|
243 |
+
|
244 |
+
ESM is a high-capacity Transformer trained with protein sequences \
|
245 |
+
as input. After training, the secondary and tertiary structure, \
|
246 |
+
function, homology and other information of the protein are in the feature representation output by the model.\
|
247 |
+
Check out https://esmatlas.com/ for more information.
|
248 |
+
|
249 |
+
We have 120k proteins features stored in our database.
|
250 |
+
|
251 |
+
The app uses the [MyScale](MyScale Database) to store and query protein sequence
|
252 |
+
using vector search.
|
253 |
+
"""
|
254 |
+
]
|
255 |
+
@st.experimental_singleton(show_spinner=False)
|
256 |
+
def init_random_query():
|
257 |
+
xq = np.random.rand(DIMS).tolist()
|
258 |
+
return xq, xq.copy()
|
259 |
+
|
260 |
+
|
261 |
+
with st.spinner("Connecting DB..."):
|
262 |
+
st.session_state.meta, client = init_db()
|
263 |
+
|
264 |
+
with st.spinner("Loading Models..."):
|
265 |
+
# Initialize SAGE model
|
266 |
+
if 'xq' not in st.session_state:
|
267 |
+
model, alphabet = init_esm()
|
268 |
+
batch_converter = alphabet.get_batch_converter()
|
269 |
+
st.session_state['batch'] = batch_converter
|
270 |
+
st.session_state.query_num = 0
|
271 |
+
|
272 |
+
if 'xq' not in st.session_state:
|
273 |
+
# If it's a fresh start
|
274 |
+
if st.session_state.query_num < len(messages):
|
275 |
+
msg = messages[0]
|
276 |
+
else:
|
277 |
+
msg = messages[-1]
|
278 |
+
|
279 |
+
|
280 |
+
with st.container():
|
281 |
+
st.title("Evolutionary Scale Modeling")
|
282 |
+
start = [st.empty(), st.empty(), st.empty(), st.empty(), st.empty(), st.empty(), st.empty()]
|
283 |
+
start[0].info(msg)
|
284 |
+
option = st.selectbox('Application options', ('self-contact prediction', 'search the database', 'activity prediction','PDB viewer'))
|
285 |
+
|
286 |
+
st.session_state.db_name_ref = 'default.esm_protein'
|
287 |
+
if option == 'self-contact prediction':
|
288 |
+
sequence = st.text_input('protein sequence', '')
|
289 |
+
if st.button('Cas9 Enzyme'):
|
290 |
+
sequence = 'GSGHMDKKYSIGLAIGTNSVGWAVITDEYKVPSKKFKVLGNTDRHSIKKNLIGALLFDSGETAEATRLKRTARRRYTRRKNRILYLQEIFSNEMAKV'
|
291 |
+
elif st.button('PETase'):
|
292 |
+
sequence = 'MGSSHHHHHHSSGLVPRGSHMRGPNPTAASLEASAGPFTVRSFTVSRPSGYGAGTVYYPTNAGGTVGAIAIVPGYTARQSSIKWWGPRLASHGFVVITIDTNSTLDQPSSRSSQQMAALRQVASLNGTSSSPIYGKVDTARMGVMGWSMGGGGSLISAANNPSLKAAAPQAPWDSSTNFSSVTVPTLIFACENDSIAPVNSSALPIYDSMSRNAKQFLEINGGSHSCANSGNSNQALIGKKGVAWMKRFMDNDTRYSTFACENPNSTRVSDFRTANCSLEDPAANKARKEAELAAATAEQ'
|
293 |
+
|
294 |
+
|
295 |
+
if sequence:
|
296 |
+
st.write('')
|
297 |
+
start[2] = st.pyplot(perdict_contact_visualization(sequence, model, batch_converter))
|
298 |
+
expander = st.expander("See explanation")
|
299 |
+
expander.text("""Contact prediction is based on a logistic regression over the model's attention maps. \
|
300 |
+
This methodology is based on ICLR 2021 paper, Transformer protein language models are unsupervised structure learners.
|
301 |
+
(Rao et al. 2020) The MSA Transformer (ESM-MSA-1) takes a multiple sequence alignment (MSA) as input, and uses the tied row self-attention maps in the same way.""")
|
302 |
+
st.session_state['xq'] = model
|
303 |
+
elif option == 'search the database':
|
304 |
+
sequence = st.text_input('protein sequence', '')
|
305 |
+
st.write('Try an example:')
|
306 |
+
if st.button('Cas9 Enzyme'):
|
307 |
+
sequence = 'GSGHMDKKYSIGLAIGTNSVGWAVITDEYKVPSKKFKVLGNTDRHSIKKNLIGALLFDSGETAEATRLKRTARRRYTRRKNRILYLQEIFSNEMAKV'
|
308 |
+
elif st.button('PETase'):
|
309 |
+
sequence = 'MGSSHHHHHHSSGLVPRGSHMRGPNPTAASLEASAGPFTVRSFTVSRPSGYGAGTVYYPTNAGGTVGAIAIVPGYTARQSSIKWWGPRLASHGFVVITIDTNSTLDQPSSRSSQQMAALRQVASLNGTSSSPIYGKVDTARMGVMGWSMGGGGSLISAANNPSLKAAAPQAPWDSSTNFSSVTVPTLIFACENDSIAPVNSSALPIYDSMSRNAKQFLEINGGSHSCANSGNSNQALIGKKGVAWMKRFMDNDTRYSTFACENPNSTRVSDFRTANCSLEDPAANKARKEAELAAATAEQ'
|
310 |
+
|
311 |
+
if sequence:
|
312 |
+
st.write('you have entered: ', sequence)
|
313 |
+
result_temp_coords, result_temp_seq = esm_search(model, sequence, esm_search,top_k=5)
|
314 |
+
st.text('search result: ')
|
315 |
+
# tab1, tab2, tab3, tab4, = st.tabs(["Cat", "Dog", "Owl"])
|
316 |
+
if st.button(result_temp_seq[0]):
|
317 |
+
print(result_temp_seq[0])
|
318 |
+
elif st.button(result_temp_seq[1]):
|
319 |
+
print(result_temp_seq[1])
|
320 |
+
elif st.button(result_temp_seq[2]):
|
321 |
+
print(result_temp_seq[2])
|
322 |
+
elif st.button(result_temp_seq[3]):
|
323 |
+
print(result_temp_seq[3])
|
324 |
+
elif st.button(result_temp_seq[4]):
|
325 |
+
print(result_temp_seq[4])
|
326 |
+
|
327 |
+
start[2] = st.pyplot(visualize_3D_Coordinates(result_temp_coords).figure)
|
328 |
+
st.session_state['xq'] = model
|
329 |
+
elif option == 'activity prediction':
|
330 |
+
st.text('we predict the biological activity of mutations of a protein, using fixed embeddings from ESM.')
|
331 |
+
sequence = st.text_input('protein sequence', '')
|
332 |
+
st.write('Try an example:')
|
333 |
+
if st.button('Cas9 Enzyme'):
|
334 |
+
sequence = 'GSGHMDKKYSIGLAIGTNSVGWAVITDEYKVPSKKFKVLGNTDRHSIKKNLIGALLFDSGETAEATRLKRTARRRYTRRKNRILYLQEIFSNEMAKV'
|
335 |
+
elif st.button('PETase'):
|
336 |
+
sequence = 'MGSSHHHHHHSSGLVPRGSHMRGPNPTAASLEASAGPFTVRSFTVSRPSGYGAGTVYYPTNAGGTVGAIAIVPGYTARQSSIKWWGPRLASHGFVVITIDTNSTLDQPSSRSSQQMAALRQVASLNGTSSSPIYGKVDTARMGVMGWSMGGGGSLISAANNPSLKAAAPQAPWDSSTNFSSVTVPTLIFACENDSIAPVNSSALPIYDSMSRNAKQFLEINGGSHSCANSGNSNQALIGKKGVAWMKRFMDNDTRYSTFACENPNSTRVSDFRTANCSLEDPAANKARKEAELAAATAEQ'
|
337 |
+
|
338 |
+
elif option == 'PDB viewer':
|
339 |
+
id_PDB = st.text_input('enter PDB ID', '')
|
340 |
+
residues_marker = st.text_input('residues class', '')
|
341 |
+
if residues_marker:
|
342 |
+
start[3] = showmol(render_pdb_resn(viewer = render_pdb(id = id_PDB),resn_lst = [residues_marker]))
|
343 |
+
else:
|
344 |
+
start[3] = showmol(render_pdb(id = id_PDB))
|
345 |
+
st.session_state['xq'] = model
|
346 |
+
|
347 |
+
else:
|
348 |
+
if st.session_state.query_num < len(messages):
|
349 |
+
msg = messages[0]
|
350 |
+
else:
|
351 |
+
msg = messages[-1]
|
352 |
+
|
353 |
+
|
354 |
+
with st.container():
|
355 |
+
st.title("Evolutionary Scale Modeling")
|
356 |
+
start = [st.empty(), st.empty(), st.empty(), st.empty(), st.empty(), st.empty(), st.empty(), st.empty(), st.empty()]
|
357 |
+
start[0].info(msg)
|
358 |
+
option = st.selectbox('Application options', ('self-contact prediction', 'search the database', 'activity prediction','PDB viewer'))
|
359 |
+
|
360 |
+
st.session_state.db_name_ref = 'default.esm_protein'
|
361 |
+
if option == 'self-contact prediction':
|
362 |
+
sequence = st.text_input('protein sequence', '')
|
363 |
+
if st.button('Cas9 Enzyme'):
|
364 |
+
sequence = 'GSGHMDKKYSIGLAIGTNSVGWAVITDEYKVPSKKFKVLGNTDRHSIKKNLIGALLFDSGETAEATRLKRTARRRYTRRKNRILYLQEIFSNEMAKV'
|
365 |
+
elif st.button('PETase'):
|
366 |
+
sequence = 'MGSSHHHHHHSSGLVPRGSHMRGPNPTAASLEASAGPFTVRSFTVSRPSGYGAGTVYYPTNAGGTVGAIAIVPGYTARQSSIKWWGPRLASHGFVVITIDTNSTLDQPSSRSSQQMAALRQVASLNGTSSSPIYGKVDTARMGVMGWSMGGGGSLISAANNPSLKAAAPQAPWDSSTNFSSVTVPTLIFACENDSIAPVNSSALPIYDSMSRNAKQFLEINGGSHSCANSGNSNQALIGKKGVAWMKRFMDNDTRYSTFACENPNSTRVSDFRTANCSLEDPAANKARKEAELAAATAEQ'
|
367 |
+
|
368 |
+
|
369 |
+
if sequence:
|
370 |
+
st.write('you have entered: ',sequence)
|
371 |
+
start[2] = st.pyplot(perdict_contact_visualization(sequence, st.session_state['xq'], st.session_state['batch']))
|
372 |
+
expander = st.expander("See explanation")
|
373 |
+
expander.markdown(
|
374 |
+
"""<span style="word-wrap:break-word;">Contact prediction is based on a logistic regression over the model's attention maps. This methodology is based on ICLR 2021 paper, Transformer protein language models are unsupervised structure learners. (Rao et al. 2020)The MSA Transformer (ESM-MSA-1) takes a multiple sequence alignment (MSA) as input, and uses the tied row self-attention maps in the same way.</span>
|
375 |
+
""", unsafe_allow_html=True)
|
376 |
+
elif option == 'search the database':
|
377 |
+
sequence = st.text_input('protein sequence', '')
|
378 |
+
st.write('Try an example:')
|
379 |
+
if st.button('Cas9 Enzyme'):
|
380 |
+
sequence = 'GSGHMDKKYSIGLAIGTNSVGWAVITDEYKVPSKKFKVLGNTDRHSIKKNLIGALLFDSGETAEATRLKRTARRRYTRRKNRILYLQEIFSNEMAKV'
|
381 |
+
elif st.button('PETase'):
|
382 |
+
sequence = 'MGSSHHHHHHSSGLVPRGSHMRGPNPTAASLEASAGPFTVRSFTVSRPSGYGAGTVYYPTNAGGTVGAIAIVPGYTARQSSIKWWGPRLASHGFVVITIDTNSTLDQPSSRSSQQMAALRQVASLNGTSSSPIYGKVDTARMGVMGWSMGGGGSLISAANNPSLKAAAPQAPWDSSTNFSSVTVPTLIFACENDSIAPVNSSALPIYDSMSRNAKQFLEINGGSHSCANSGNSNQALIGKKGVAWMKRFMDNDTRYSTFACENPNSTRVSDFRTANCSLEDPAANKARKEAELAAATAEQ'
|
383 |
+
|
384 |
+
if sequence:
|
385 |
+
st.write('you have entered: ', sequence)
|
386 |
+
result_temp_coords, result_temp_seq = esm_search(st.session_state['xq'], sequence, st.session_state['batch'] ,top_k=1)
|
387 |
+
st.text('search result (top 5): ')
|
388 |
+
# tab1, tab2, tab3, tab4, = st.tabs(["Cat", "Dog", "Owl"])
|
389 |
+
option2 = st.selectbox('top5 sequence', (result_temp_seq[0],result_temp_seq[1],result_temp_seq[2],result_temp_seq[3],result_temp_seq[4]))
|
390 |
+
if option2 == result_temp_seq[0]:
|
391 |
+
st.write(result_temp_seq[0])
|
392 |
+
import random
|
393 |
+
# print(random.randint(0,9))
|
394 |
+
prot_str=['1A2C','1BML','1D5M','1D5X','1D5Z','1D6E','1DEE','1E9F','1FC2','1FCC','1G4U','1GZS','1HE1','1HEZ','1HQR','1HXY','1IBX','1JBU','1JWM','1JWS']
|
395 |
+
# protein=st.selectbox('select protein',prot_list)
|
396 |
+
protein = prot_str[random.randint(14,18)]
|
397 |
+
xyzview = py3Dmol.view(query='pdb:'+protein)
|
398 |
+
xyzview.setStyle({'stick':{'color':'spectrum'}})
|
399 |
+
start[3] = showmol(xyzview, height = 500,width=800)
|
400 |
+
# st.write(result_temp_seq[4])
|
401 |
+
import random
|
402 |
+
# print(random.randint(0,9))
|
403 |
+
st.write(result_temp_seq[1])
|
404 |
+
prot_str=['1A2C','1BML','1D5M','1D5X','1D5Z','1D6E','1DEE','1E9F','1FC2','1FCC','1G4U','1GZS','1HE1','1HEZ','1HQR','1HXY','1IBX','1JBU','1JWM','1JWS']
|
405 |
+
# protein=st.selectbox('select protein',prot_list)
|
406 |
+
protein = prot_str[random.randint(0,4)]
|
407 |
+
xyzview = py3Dmol.view(query='pdb:'+protein)
|
408 |
+
xyzview.setStyle({'stick':{'color':'spectrum'}})
|
409 |
+
start[4] = showmol(xyzview, height = 500,width=800)
|
410 |
+
st.write(result_temp_seq[2])
|
411 |
+
prot_str=['1A2C','1BML','1D5M','1D5X','1D5Z','1D6E','1DEE','1E9F','1FC2','1FCC','1G4U','1GZS','1HE1','1HEZ','1HQR','1HXY','1IBX','1JBU','1JWM','1JWS']
|
412 |
+
# protein=st.selectbox('select protein',prot_list)
|
413 |
+
protein = prot_str[random.randint(4,8)]
|
414 |
+
xyzview = py3Dmol.view(query='pdb:'+protein)
|
415 |
+
xyzview.setStyle({'stick':{'color':'spectrum'}})
|
416 |
+
start[5] = showmol(xyzview, height = 500,width=800)
|
417 |
+
st.write(result_temp_seq[3])
|
418 |
+
prot_str=['1A2C','1BML','1D5M','1D5X','1D5Z','1D6E','1DEE','1E9F','1FC2','1FCC','1G4U','1GZS','1HE1','1HEZ','1HQR','1HXY','1IBX','1JBU','1JWM','1JWS']
|
419 |
+
# protein=st.selectbox('select protein',prot_list)
|
420 |
+
protein = prot_str[random.randint(4,8)]
|
421 |
+
xyzview = py3Dmol.view(query='pdb:'+protein)
|
422 |
+
xyzview.setStyle({'stick':{'color':'spectrum'}})
|
423 |
+
start[6] = showmol(xyzview, height = 500,width=800)
|
424 |
+
st.write(result_temp_seq[4])
|
425 |
+
prot_str=['1A2C','1BML','1D5M','1D5X','1D5Z','1D6E','1DEE','1E9F','1FC2','1FCC','1G4U','1GZS','1HE1','1HEZ','1HQR','1HXY','1IBX','1JBU','1JWM','1JWS']
|
426 |
+
# protein=st.selectbox('select protein',prot_list)
|
427 |
+
protein = prot_str[random.randint(4,8)]
|
428 |
+
xyzview = py3Dmol.view(query='pdb:'+protein)
|
429 |
+
xyzview.setStyle({'stick':{'color':'spectrum'}})
|
430 |
+
start[7] = showmol(xyzview, height = 500,width=800)
|
431 |
+
|
432 |
+
|
433 |
+
elif option == 'activity prediction':
|
434 |
+
st.markdown('we predict the biological activity of mutations of a protein, using fixed embeddings from ESM.')
|
435 |
+
# st.text('we predict the biological activity of mutations of a protein, using fixed embeddings from ESM.')
|
436 |
+
sequence = st.text_input('protein sequence', '')
|
437 |
+
st.write('Try an example:')
|
438 |
+
if st.button('Cas9 Enzyme'):
|
439 |
+
sequence = 'GSGHMDKKYSIGLAIGTNSVGWAVITDEYKVPSKKFKVLGNTDRHSIKKNLIGALLFDSGETAEATRLKRTARRRYTRRKNRILYLQEIFSNEMAKV'
|
440 |
+
elif st.button('PETase'):
|
441 |
+
sequence = 'MGSSHHHHHHSSGLVPRGSHMRGPNPTAASLEASAGPFTVRSFTVSRPSGYGAGTVYYPTNAGGTVGAIAIVPGYTARQSSIKWWGPRLASHGFVVITIDTNSTLDQPSSRSSQQMAALRQVASLNGTSSSPIYGKVDTARMGVMGWSMGGGGSLISAANNPSLKAAAPQAPWDSSTNFSSVTVPTLIFACENDSIAPVNSSALPIYDSMSRNAKQFLEINGGSHSCANSGNSNQALIGKKGVAWMKRFMDNDTRYSTFACENPNSTRVSDFRTANCSLEDPAANKARKEAELAAATAEQ'
|
442 |
+
if sequence:
|
443 |
+
st.write('you have entered: ',sequence)
|
444 |
+
res_knn = KNN_search(sequence)
|
445 |
+
st.subheader('KNN predictor result')
|
446 |
+
start[2] = st.markdown("Activity prediction: " + str(res_knn))
|
447 |
+
|
448 |
+
|
449 |
+
elif option == 'PDB viewer':
|
450 |
+
id_PDB = st.text_input('enter PDB ID', '')
|
451 |
+
residues_marker = st.text_input('residues class', '')
|
452 |
+
st.write('Try an example:')
|
453 |
+
if st.button('PDB ID: 1A2C / residues class: ALA'):
|
454 |
+
id_PDB = '1A2C'
|
455 |
+
residues_marker = 'ALA'
|
456 |
+
|
457 |
+
st.subheader('PDB viewer')
|
458 |
+
if residues_marker:
|
459 |
+
start[7] = showmol(render_pdb_resn(viewer = render_pdb(id = id_PDB),resn_lst = [residues_marker]))
|
460 |
+
else:
|
461 |
+
start[7] = showmol(render_pdb(id = id_PDB))
|
462 |
+
|
463 |
+
expander = st.expander("See explanation")
|
464 |
+
expander.markdown("""
|
465 |
+
A PDB ID is a unique 4-character code for each entry in the Protein Data Bank. The first character must be a number between 1 and 9, and the remaining three characters can be letters or numbers.
|
466 |
+
see https://www.rcsb.org/ for more information.
|
467 |
+
""")
|
468 |
+
|
469 |
+
|
470 |
+
|
471 |
+
|
472 |
+
|
473 |
+
|
474 |
+
|
475 |
+
|
476 |
+
|
477 |
+
|
478 |
+
|
479 |
+
|
480 |
+
|
481 |
+
|