File size: 7,067 Bytes
f07511a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
# -*- coding: utf-8 -*-
"""Untitled17.ipynb

Automatically generated by Colab.

Original file is located at
    https://colab.research.google.com/drive/1GwdSjrwh3f6QCzOa8KHr_XkWy0KmZvdV
"""

import pandas as pd
import numpy as np
from pathlib import Path
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import roc_auc_score, average_precision_score, roc_curve, precision_recall_curve
from sklearn.linear_model import LogisticRegression
import matplotlib.pyplot as plt
from tqdm import tqdm
import pickle
import torch
import torch.nn as nn
import warnings
warnings.filterwarnings('ignore')


PATHS = {
    'features': BASE_PATH / 'features',
    'embeddings': BASE_PATH / 'embeddings',
    'models': BASE_PATH / 'models',
    'results': BASE_PATH / 'results',
    'figures': BASE_PATH / 'results' / 'figures',
}

df_features = pd.read_parquet(PATHS['features'] / 'features_classical_full.parquet')

id_cols = ['mutation_idx', 'uniprot_acc', 'gene_symbol', 'position', 'wt_aa', 'mut_aa', 'label']
feature_cols = [c for c in df_features.columns if c not in id_cols]

X_features = df_features[feature_cols].values.astype(np.float32)
X_features = np.nan_to_num(X_features, nan=0.0, posinf=0.0, neginf=0.0)
y = df_features['label'].values
proteins = df_features['uniprot_acc'].values

X_emb_combined = np.load(PATHS['embeddings'] / 'embeddings_combined_full.npy').astype(np.float32)
X_emb_local = np.load(PATHS['embeddings'] / 'embeddings_local_full.npy').astype(np.float32)

print(f"    {X_features.shape}")
print(f"    {X_emb_combined.shape}")



from sklearn.decomposition import PCA

pca_combined = PCA(n_components=128, random_state=42)
X_emb_pca = pca_combined.fit_transform(X_emb_combined).astype(np.float32)

pca_local = PCA(n_components=64, random_state=42)
X_emb_local_pca = pca_local.fit_transform(X_emb_local).astype(np.float32)



configs = [
    {'name': 'Features classiques', 'X': X_features},
    {'name': 'Embeddings ESM-2', 'X': X_emb_pca},
    {'name': 'Features + Embeddings', 'X': np.concatenate([X_features, X_emb_pca], axis=1)},
    {'name': 'Features + Emb. Local', 'X': np.concatenate([X_features, X_emb_local_pca], axis=1)},
]

class SimpleMLP(nn.Module):
    def __init__(self, input_dim, hidden_dim=256):
        super().__init__()
        self.net = nn.Sequential(
            nn.Linear(input_dim, hidden_dim),
            nn.ReLU(),
            nn.Dropout(0.3),
            nn.Linear(hidden_dim, hidden_dim // 2),
            nn.ReLU(),
            nn.Dropout(0.2),
            nn.Linear(hidden_dim // 2, 1),
            nn.Sigmoid()
        )

    def forward(self, x):
        return self.net(x).squeeze()

def train_mlp(X_train, y_train, X_test, input_dim, device, epochs=50, lr=0.001):

    model = SimpleMLP(input_dim).to(device)
    optimizer = torch.optim.Adam(model.parameters(), lr=lr, weight_decay=1e-4)
    criterion = nn.BCELoss()

    X_train_t = torch.FloatTensor(X_train).to(device)
    y_train_t = torch.FloatTensor(y_train).to(device)
    X_test_t = torch.FloatTensor(X_test).to(device)

    model.train()
    batch_size = 512

    for epoch in range(epochs):
        perm = torch.randperm(len(X_train_t))

        for i in range(0, len(X_train_t), batch_size):
            idx = perm[i:i+batch_size]

            optimizer.zero_grad()
            outputs = model(X_train_t[idx])
            loss = criterion(outputs, y_train_t[idx])
            loss.backward()
            optimizer.step()

    model.eval()
    with torch.no_grad():
        y_pred = model(X_test_t).cpu().numpy()

    return y_pred

def evaluate_lpocv_gpu(X, y, proteins, device, epochs=30):

    unique_proteins = np.unique(proteins)
    results = []

    scaler = StandardScaler()
    X_scaled = scaler.fit_transform(X)

    for protein in tqdm(unique_proteins, desc="LPOCV GPU"):
        test_mask = proteins == protein
        train_mask = ~test_mask

        if test_mask.sum() < 2:
            continue

        X_train, y_train = X_scaled[train_mask], y[train_mask]
        X_test, y_test = X_scaled[test_mask], y[test_mask]

        y_pred = train_mlp(X_train, y_train, X_test, X.shape[1], device, epochs=epochs)

        for pred, true in zip(y_pred, y_test):
            results.append({'y_true': true, 'y_pred': float(pred)})

    df_res = pd.DataFrame(results)

    if len(df_res) > 0 and len(df_res['y_true'].unique()) > 1:
        auc_roc = roc_auc_score(df_res['y_true'], df_res['y_pred'])
        auc_pr = average_precision_score(df_res['y_true'], df_res['y_pred'])
    else:
        auc_roc, auc_pr = 0, 0

    return auc_roc, auc_pr, df_res

results_all = {}

for cfg in configs:
    print(f"\n   📊 {cfg['name']}...")

    auc_roc, auc_pr, df_res = evaluate_lpocv_gpu(
        cfg['X'], y, proteins, device, epochs=30
    )

    results_all[cfg['name']] = {
        'auc_roc': auc_roc,
        'auc_pr': auc_pr,
        'predictions': df_res,
        'n_features': cfg['X'].shape[1],
    }

    print(f"      AUC-ROC: {auc_roc:.4f}")
    print(f"      AUC-PR:  {auc_pr:.4f}")


def evaluate_lpocv_logreg(X, y, proteins):

    unique_proteins = np.unique(proteins)
    results = []

    scaler = StandardScaler()
    X_scaled = scaler.fit_transform(X)

    for protein in tqdm(unique_proteins, desc="LPOCV LogReg", leave=False):
        test_mask = proteins == protein
        train_mask = ~test_mask

        if test_mask.sum() < 2:
            continue

        X_train, y_train = X_scaled[train_mask], y[train_mask]
        X_test, y_test = X_scaled[test_mask], y[test_mask]

        model = LogisticRegression(max_iter=500, C=0.1, random_state=42)
        model.fit(X_train, y_train)
        y_pred = model.predict_proba(X_test)[:, 1]

        for pred, true in zip(y_pred, y_test):
            results.append({'y_true': true, 'y_pred': pred})

    df_res = pd.DataFrame(results)

    if len(df_res) > 0:
        auc_roc = roc_auc_score(df_res['y_true'], df_res['y_pred'])
        auc_pr = average_precision_score(df_res['y_true'], df_res['y_pred'])
    else:
        auc_roc, auc_pr = 0, 0

    return auc_roc, auc_pr

for cfg in configs:
    auc_roc, auc_pr = evaluate_lpocv_logreg(cfg['X'], y, proteins)
    results_all[cfg['name']]['auc_roc_logreg'] = auc_roc
    results_all[cfg['name']]['auc_pr_logreg'] = auc_pr
    print(f"   {cfg['name']}: LogReg AUC-ROC = {auc_roc:.4f}")


comparison_data = []
for name, res in results_all.items():
    comparison_data.append({
        'Configuration': name,
        'Features': res['n_features'],
        'AUC-ROC (MLP)': res['auc_roc'],
        'AUC-PR (MLP)': res['auc_pr'],
        'AUC-ROC (LogReg)': res.get('auc_roc_logreg', 0),
    })

df_comparison = pd.DataFrame(comparison_data)
df_comparison = df_comparison.sort_values('AUC-ROC (MLP)', ascending=False)

print("\n" + df_comparison.to_string(index=False))

best_name = df_comparison.iloc[0]['Configuration']
best_auc = df_comparison.iloc[0]['AUC-ROC (MLP)']
print(f"\n   🏆 Meilleur: {best_name} (AUC-ROC = {best_auc:.4f})")

df_comparison.to_csv(PATHS['results'] / 'comparison_final.csv', index=False)