Update app.py
Browse files
app.py
CHANGED
@@ -30,31 +30,35 @@ class Config:
|
|
30 |
|
31 |
|
32 |
def predict_stability_with_pdb(model_choice, organism_choice, pdb_files, cfg=Config()):
|
33 |
-
results = []
|
|
|
|
|
|
|
|
|
|
|
|
|
34 |
for pdb_file in pdb_files:
|
35 |
try:
|
36 |
pdb_path = pdb_file.name
|
37 |
os.system("chmod 777 bin/foldseek")
|
38 |
sequences = get_foldseek_seq(pdb_path)
|
39 |
if not sequences:
|
40 |
-
results.append(
|
41 |
-
|
42 |
-
|
43 |
-
})
|
44 |
continue
|
45 |
|
46 |
sequence = sequences[2] if model_choice == "SaProt" else sequences[0]
|
47 |
-
|
48 |
-
|
49 |
-
results.append({"file_name": pdb_path,
|
50 |
-
"raw prediction value": output["raw prediction values"][0],
|
51 |
-
"binary prediction value": output["binary prediction values"][0]
|
52 |
-
})
|
53 |
except Exception as e:
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
|
|
|
|
|
|
58 |
|
59 |
df = pd.DataFrame(results)
|
60 |
output_csv = "/tmp/predictions.csv"
|
@@ -72,13 +76,13 @@ def predict_stability_with_sequence(model_choice, organism_choice, sequence, cfg
|
|
72 |
return f"An error occurred: {str(e)}"
|
73 |
|
74 |
|
75 |
-
def predict_stability_core(model_choice, organism_choice,
|
76 |
cell_line = "HeLa" if organism_choice == "Human" else "NIH3T3"
|
77 |
cfg.model = f"sagawa/PLTNUM-{model_choice}-{cell_line}"
|
78 |
cfg.architecture = model_choice
|
79 |
cfg.model_path = f"sagawa/PLTNUM-{model_choice}-{cell_line}"
|
80 |
|
81 |
-
output = predict(cfg,
|
82 |
return output
|
83 |
|
84 |
|
@@ -92,7 +96,7 @@ def get_foldseek_seq(pdb_path):
|
|
92 |
return parsed_seqs
|
93 |
|
94 |
|
95 |
-
def predict(cfg,
|
96 |
cfg.token_length = 2 if cfg.architecture == "SaProt" else 1
|
97 |
cfg.device = "cuda" if torch.cuda.is_available() else "cpu"
|
98 |
|
@@ -100,7 +104,7 @@ def predict(cfg, sequence):
|
|
100 |
cfg.max_length += 1
|
101 |
|
102 |
seed_everything(cfg.seed)
|
103 |
-
df = pd.DataFrame({cfg.sequence_col:
|
104 |
|
105 |
tokenizer = AutoTokenizer.from_pretrained(
|
106 |
cfg.model_path, padding_side=cfg.padding_side
|
@@ -134,19 +138,8 @@ def predict(cfg, sequence):
|
|
134 |
predictions += preds.cpu().tolist()
|
135 |
|
136 |
predictions = list(itertools.chain.from_iterable(predictions))
|
137 |
-
outputs = {
|
138 |
-
"raw prediction values": predictions,
|
139 |
-
"binary prediction values": [1 if x > 0.5 else 0 for x in predictions]
|
140 |
-
}
|
141 |
-
|
142 |
-
html_output = f"""
|
143 |
-
<div style='border: 2px solid #4CAF50; padding: 10px; border-radius: 10px;'>
|
144 |
-
<p><strong>Raw prediction value:</strong> {outputs['raw prediction values'][0]}</p>
|
145 |
-
<p><strong>Binary prediction values:</strong> {outputs['binary prediction values'][0]}</p>
|
146 |
-
</div>
|
147 |
-
"""
|
148 |
|
149 |
-
return
|
150 |
|
151 |
|
152 |
# Gradio Interface
|
|
|
30 |
|
31 |
|
32 |
def predict_stability_with_pdb(model_choice, organism_choice, pdb_files, cfg=Config()):
|
33 |
+
results = {"file_name": [],
|
34 |
+
"raw prediction value": [],
|
35 |
+
"binary prediction value": []
|
36 |
+
}
|
37 |
+
file_names = []
|
38 |
+
sequences = []
|
39 |
+
|
40 |
for pdb_file in pdb_files:
|
41 |
try:
|
42 |
pdb_path = pdb_file.name
|
43 |
os.system("chmod 777 bin/foldseek")
|
44 |
sequences = get_foldseek_seq(pdb_path)
|
45 |
if not sequences:
|
46 |
+
results["file_name"].append(pdb_file.name)
|
47 |
+
results["raw prediction value"].append(None)
|
48 |
+
results["binary prediction value"].append(None)
|
|
|
49 |
continue
|
50 |
|
51 |
sequence = sequences[2] if model_choice == "SaProt" else sequences[0]
|
52 |
+
file_names.append(pdb_file.name)
|
53 |
+
sequences.append(sequence)
|
|
|
|
|
|
|
|
|
54 |
except Exception as e:
|
55 |
+
results["file_name"].append(pdb_file.name)
|
56 |
+
results["raw prediction value"].append(None)
|
57 |
+
results["binary prediction value"].append(None)
|
58 |
+
raw_prediction, binary_prediction = predict_stability_core(model_choice, organism_choice, sequences, cfg)
|
59 |
+
results["file_name"] = results["file_name"] + file_names
|
60 |
+
results["raw prediction value"] = results["raw prediction value"] + raw_prediction
|
61 |
+
results["binary prediction value"] = results["binary prediction value"] + binary_prediction
|
62 |
|
63 |
df = pd.DataFrame(results)
|
64 |
output_csv = "/tmp/predictions.csv"
|
|
|
76 |
return f"An error occurred: {str(e)}"
|
77 |
|
78 |
|
79 |
+
def predict_stability_core(model_choice, organism_choice, sequences, cfg=Config()):
|
80 |
cell_line = "HeLa" if organism_choice == "Human" else "NIH3T3"
|
81 |
cfg.model = f"sagawa/PLTNUM-{model_choice}-{cell_line}"
|
82 |
cfg.architecture = model_choice
|
83 |
cfg.model_path = f"sagawa/PLTNUM-{model_choice}-{cell_line}"
|
84 |
|
85 |
+
output = predict(cfg, sequences)
|
86 |
return output
|
87 |
|
88 |
|
|
|
96 |
return parsed_seqs
|
97 |
|
98 |
|
99 |
+
def predict(cfg, sequences):
|
100 |
cfg.token_length = 2 if cfg.architecture == "SaProt" else 1
|
101 |
cfg.device = "cuda" if torch.cuda.is_available() else "cpu"
|
102 |
|
|
|
104 |
cfg.max_length += 1
|
105 |
|
106 |
seed_everything(cfg.seed)
|
107 |
+
df = pd.DataFrame({cfg.sequence_col: sequences})
|
108 |
|
109 |
tokenizer = AutoTokenizer.from_pretrained(
|
110 |
cfg.model_path, padding_side=cfg.padding_side
|
|
|
138 |
predictions += preds.cpu().tolist()
|
139 |
|
140 |
predictions = list(itertools.chain.from_iterable(predictions))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
141 |
|
142 |
+
return predictions, [1 if x > 0.5 else 0 for x in predictions]
|
143 |
|
144 |
|
145 |
# Gradio Interface
|