Creeperbbs commited on
Commit
cf90968
·
verified ·
1 Parent(s): a0b8fcc

Update RMMol release bundle v1.1

Browse files
Files changed (4) hide show
  1. MODEL_CARD.md +3 -3
  2. README.md +4 -4
  3. dataset/README.md +1 -1
  4. examples/RMMol_sample.ipynb +606 -0
MODEL_CARD.md CHANGED
@@ -7,12 +7,12 @@ tags:
7
  - molecular-representation
8
  - rmmol
9
  - frozen-embeddings
10
- - molecule-benchmark
11
  ---
12
 
13
  # RMMol
14
 
15
- RMMol is a frozen molecular representation model based on a geometry-aware decoupled message passing encoder. This release bundle provides standardized benchmark CSVs and a RMMol-only Jupyter sample for four domains: Biophysics, Physiological, Physical Chemistry, and Quantum Mechanics.
16
 
17
  ## Weights
18
 
@@ -35,6 +35,6 @@ The current inference and downstream-probe reproduction was run on the Spark wor
35
 
36
  ## Reproduction
37
 
38
- Use `examples/rmmol_benchmark_sample.ipynb` to inspect the exact RMMol-only dataset names and selected probe settings.
39
 
40
  Generated on 2026-09-14.
 
7
  - molecular-representation
8
  - rmmol
9
  - frozen-embeddings
10
+ - molecule-representation
11
  ---
12
 
13
  # RMMol
14
 
15
+ RMMol is a frozen molecular representation model based on a geometry-aware decoupled message passing encoder. This release bundle provides standardized CSVs and a RMMol-only Jupyter sample for four domains: Biophysics, Physiological, Physical Chemistry, and Quantum Mechanics.
16
 
17
  ## Weights
18
 
 
35
 
36
  ## Reproduction
37
 
38
+ Use `examples/RMMol_sample.ipynb` to inspect the exact RMMol-only dataset names and selected probe settings.
39
 
40
  Generated on 2026-09-14.
README.md CHANGED
@@ -10,12 +10,12 @@ tags:
10
  - qm9
11
  - tox21
12
  - sider
13
- pretty_name: RMMol benchmark dataset
14
  ---
15
 
16
- # RMMol benchmark dataset
17
 
18
- This bundle contains standardized CSV files and a Jupyter sample notebook for the RMMol frozen-embedding benchmark across Biophysics, Physiological, Physical Chemistry, and Quantum Mechanics.
19
 
20
  ## Contents
21
 
@@ -23,7 +23,7 @@ This bundle contains standardized CSV files and a Jupyter sample notebook for th
23
  - `dataset/physiological/`: BACE, ClinTox FDA-approved, Tox21 single-label splits, and SIDER single-label splits.
24
  - `dataset/physical_chemistry/`: ESOL, FreeSolv, and Lipophilicity splits.
25
  - `dataset/quantum_mechanics/`: QM9 splits.
26
- - `examples/rmmol_benchmark_sample.ipynb`: RMMol-only sample notebook with the selected downstream-head settings.
27
  - `MANIFEST.csv`: file-level provenance, row counts, columns, and checksums.
28
 
29
  ## Naming convention
 
10
  - qm9
11
  - tox21
12
  - sider
13
+ pretty_name: RMMol
14
  ---
15
 
16
+ # RMMol
17
 
18
+ This bundle contains standardized CSV files and a Jupyter sample notebook for RMMol frozen-embedding examples across Biophysics, Physiological, Physical Chemistry, and Quantum Mechanics.
19
 
20
  ## Contents
21
 
 
23
  - `dataset/physiological/`: BACE, ClinTox FDA-approved, Tox21 single-label splits, and SIDER single-label splits.
24
  - `dataset/physical_chemistry/`: ESOL, FreeSolv, and Lipophilicity splits.
25
  - `dataset/quantum_mechanics/`: QM9 splits.
26
+ - `examples/RMMol_sample.ipynb`: RMMol-only sample notebook with the selected downstream-head settings.
27
  - `MANIFEST.csv`: file-level provenance, row counts, columns, and checksums.
28
 
29
  ## Naming convention
dataset/README.md CHANGED
@@ -1,5 +1,5 @@
1
  # Dataset layout
2
 
3
- This directory stores only CSV data required for the RMMol benchmark sample. Each CSV keeps the original task label columns while standardizing split names to `train`, `val`, and `test`.
4
 
5
  Biophysics contains `biophysics/bbbp/bbbp.csv` and `biophysics/hiv/hiv_{train,val,test}.csv`.
 
1
  # Dataset layout
2
 
3
+ This directory stores CSV data required for the RMMol sample. Each CSV keeps the original task label columns while standardizing split names to `train`, `val`, and `test`.
4
 
5
  Biophysics contains `biophysics/bbbp/bbbp.csv` and `biophysics/hiv/hiv_{train,val,test}.csv`.
examples/RMMol_sample.ipynb ADDED
@@ -0,0 +1,606 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "cells": [
3
+ {
4
+ "cell_type": "markdown",
5
+ "metadata": {},
6
+ "source": [
7
+ "# RMMol frozen-embedding sample\n",
8
+ "\n",
9
+ "This notebook demonstrates how to load the packaged CSV files, generate RMMol embeddings, and apply the selected downstream heads. It is RMMol-only and does not include baseline model artifacts or local output tables."
10
+ ]
11
+ },
12
+ {
13
+ "cell_type": "code",
14
+ "execution_count": null,
15
+ "metadata": {},
16
+ "outputs": [],
17
+ "source": [
18
+ "from pathlib import Path\n",
19
+ "import os, sys, json\n",
20
+ "import numpy as np\n",
21
+ "import pandas as pd\n",
22
+ "from sklearn.ensemble import ExtraTreesClassifier\n",
23
+ "from sklearn.linear_model import Ridge, RidgeClassifier\n",
24
+ "from sklearn.metrics import mean_absolute_error, mean_squared_error, roc_auc_score\n",
25
+ "from sklearn.model_selection import KFold\n",
26
+ "from sklearn.pipeline import make_pipeline\n",
27
+ "from sklearn.preprocessing import Normalizer, StandardScaler\n",
28
+ "\n",
29
+ "RELEASE_DIR = Path(os.environ.get('RMMOL_RELEASE_DIR', Path.cwd()))\n",
30
+ "if not (RELEASE_DIR / 'dataset').exists():\n",
31
+ " RELEASE_DIR = Path.cwd().parent\n",
32
+ "RMMOL_CKPT = os.environ.get('RMMOL_CKPT', '')\n",
33
+ "RMMOL_TRUE_LOADER_DIR = Path(os.environ.get('RMMOL_TRUE_LOADER_DIR', '/home/sustech_spark/project/virtual cell/FUND/experiments'))\n",
34
+ "DEVICE = os.environ.get('RMMOL_DEVICE', 'cuda')\n",
35
+ "BATCH_SIZE = int(os.environ.get('RMMOL_BATCH_SIZE', '128'))\n"
36
+ ]
37
+ },
38
+ {
39
+ "cell_type": "code",
40
+ "execution_count": null,
41
+ "metadata": {},
42
+ "outputs": [],
43
+ "source": [
44
+ "TASK_SETTINGS = {\n",
45
+ " \"biophysics\": {\n",
46
+ " \"bbbp\": \"StandardScaler + RidgeClassifier(alpha=1.0, class_weight=None, solver='lsqr'); five-fold KFold(shuffle=True, random_state=42).\",\n",
47
+ " \"hiv\": \"Normalizer + RidgeClassifier(alpha=1.6788e-05, class_weight='balanced', solver='lsqr'); train+val fit, test report.\"\n",
48
+ " },\n",
49
+ " \"physiological\": {\n",
50
+ " \"bace\": \"ExtraTreesClassifier(n_estimators=500, criterion='entropy', max_features='sqrt', class_weight='balanced').\",\n",
51
+ " \"clintox_FDA_APPROVED\": \"l2-normalized RidgeClassifier(alpha=0.0003, class_weight=None).\",\n",
52
+ " \"tox21_and_sider\": \"Task-specific RidgeClassifier presets are embedded in the Jupyter sample notebook.\"\n",
53
+ " },\n",
54
+ " \"physical_chemistry\": {\n",
55
+ " \"esol\": \"Ridge(alpha=0.003), RMSE.\",\n",
56
+ " \"freesolv\": \"Ridge(alpha=0.003), RMSE.\",\n",
57
+ " \"lipo\": \"Ridge(alpha=0.001), RMSE.\"\n",
58
+ " },\n",
59
+ " \"quantum_mechanics\": {\n",
60
+ " \"qm9\": \"StandardScaler + Ridge(alpha=0.01), MAE over 12 QM9 targets.\"\n",
61
+ " }\n",
62
+ "}\n",
63
+ "\n",
64
+ "PHYSIOLOGICAL_TASK_PRESETS = [\n",
65
+ " {\n",
66
+ " \"dataset_key\": \"tox21\",\n",
67
+ " \"task\": \"nr-ahr\",\n",
68
+ " \"preset\": \"l2_ridge_a0.0003_cwbalanced\",\n",
69
+ " \"preset_params\": {\n",
70
+ " \"alpha\": 0.0003,\n",
71
+ " \"class_weight\": \"balanced\",\n",
72
+ " \"head\": \"l2+RidgeClassifier\"\n",
73
+ " }\n",
74
+ " },\n",
75
+ " {\n",
76
+ " \"dataset_key\": \"tox21\",\n",
77
+ " \"task\": \"nr-ar-lbd\",\n",
78
+ " \"preset\": \"std_l2_ridge_a0.003_cwnone\",\n",
79
+ " \"preset_params\": {\n",
80
+ " \"alpha\": 0.003,\n",
81
+ " \"class_weight\": null,\n",
82
+ " \"head\": \"std_l2+RidgeClassifier\"\n",
83
+ " }\n",
84
+ " },\n",
85
+ " {\n",
86
+ " \"dataset_key\": \"tox21\",\n",
87
+ " \"task\": \"nr-ar\",\n",
88
+ " \"preset\": \"pca64_ridge_a100_cwbalanced\",\n",
89
+ " \"preset_params\": {\n",
90
+ " \"alpha\": 100,\n",
91
+ " \"class_weight\": \"balanced\",\n",
92
+ " \"head\": \"pca+RidgeClassifier\",\n",
93
+ " \"n_components\": 64\n",
94
+ " }\n",
95
+ " },\n",
96
+ " {\n",
97
+ " \"dataset_key\": \"tox21\",\n",
98
+ " \"task\": \"nr-aromatase\",\n",
99
+ " \"preset\": \"std_l2_ridge_a10_cwbalanced\",\n",
100
+ " \"preset_params\": {\n",
101
+ " \"alpha\": 10,\n",
102
+ " \"class_weight\": \"balanced\",\n",
103
+ " \"head\": \"std_l2+RidgeClassifier\"\n",
104
+ " }\n",
105
+ " },\n",
106
+ " {\n",
107
+ " \"dataset_key\": \"tox21\",\n",
108
+ " \"task\": \"nr-er-lbd\",\n",
109
+ " \"preset\": \"l2_ridge_a0.0003_cwbalanced\",\n",
110
+ " \"preset_params\": {\n",
111
+ " \"alpha\": 0.0003,\n",
112
+ " \"class_weight\": \"balanced\",\n",
113
+ " \"head\": \"l2+RidgeClassifier\"\n",
114
+ " }\n",
115
+ " },\n",
116
+ " {\n",
117
+ " \"dataset_key\": \"tox21\",\n",
118
+ " \"task\": \"nr-er\",\n",
119
+ " \"preset\": \"std_l2_ridge_a0.0003_cwnone\",\n",
120
+ " \"preset_params\": {\n",
121
+ " \"alpha\": 0.0003,\n",
122
+ " \"class_weight\": null,\n",
123
+ " \"head\": \"std_l2+RidgeClassifier\"\n",
124
+ " }\n",
125
+ " },\n",
126
+ " {\n",
127
+ " \"dataset_key\": \"tox21\",\n",
128
+ " \"task\": \"nr-ppar-gamma\",\n",
129
+ " \"preset\": \"std_l2_ridge_a0.003_cwnone\",\n",
130
+ " \"preset_params\": {\n",
131
+ " \"alpha\": 0.003,\n",
132
+ " \"class_weight\": null,\n",
133
+ " \"head\": \"std_l2+RidgeClassifier\"\n",
134
+ " }\n",
135
+ " },\n",
136
+ " {\n",
137
+ " \"dataset_key\": \"tox21\",\n",
138
+ " \"task\": \"sr-are\",\n",
139
+ " \"preset\": \"std_l2_ridge_a0.003_cwbalanced\",\n",
140
+ " \"preset_params\": {\n",
141
+ " \"alpha\": 0.003,\n",
142
+ " \"class_weight\": \"balanced\",\n",
143
+ " \"head\": \"std_l2+RidgeClassifier\"\n",
144
+ " }\n",
145
+ " },\n",
146
+ " {\n",
147
+ " \"dataset_key\": \"tox21\",\n",
148
+ " \"task\": \"sr-atad5\",\n",
149
+ " \"preset\": \"std_l2_ridge_a0.001_cwbalanced\",\n",
150
+ " \"preset_params\": {\n",
151
+ " \"alpha\": 0.001,\n",
152
+ " \"class_weight\": \"balanced\",\n",
153
+ " \"head\": \"std_l2+RidgeClassifier\"\n",
154
+ " }\n",
155
+ " },\n",
156
+ " {\n",
157
+ " \"dataset_key\": \"tox21\",\n",
158
+ " \"task\": \"sr-hse\",\n",
159
+ " \"preset\": \"l2_ridge_a3e-05_cwnone\",\n",
160
+ " \"preset_params\": {\n",
161
+ " \"alpha\": 3e-05,\n",
162
+ " \"class_weight\": null,\n",
163
+ " \"head\": \"l2+RidgeClassifier\"\n",
164
+ " }\n",
165
+ " },\n",
166
+ " {\n",
167
+ " \"dataset_key\": \"tox21\",\n",
168
+ " \"task\": \"sr-mmp\",\n",
169
+ " \"preset\": \"std_l2_ridge_a0.001_cwnone\",\n",
170
+ " \"preset_params\": {\n",
171
+ " \"alpha\": 0.001,\n",
172
+ " \"class_weight\": null,\n",
173
+ " \"head\": \"std_l2+RidgeClassifier\"\n",
174
+ " }\n",
175
+ " },\n",
176
+ " {\n",
177
+ " \"dataset_key\": \"tox21\",\n",
178
+ " \"task\": \"sr-p53\",\n",
179
+ " \"preset\": \"std_l2_ridge_a1e-05_cwbalanced\",\n",
180
+ " \"preset_params\": {\n",
181
+ " \"alpha\": 1e-05,\n",
182
+ " \"class_weight\": \"balanced\",\n",
183
+ " \"head\": \"std_l2+RidgeClassifier\"\n",
184
+ " }\n",
185
+ " },\n",
186
+ " {\n",
187
+ " \"dataset_key\": \"sider\",\n",
188
+ " \"task\": \"blood and lymphatic system disorders\",\n",
189
+ " \"preset\": \"l2_ridge_a0.3_cwbalanced\",\n",
190
+ " \"preset_params\": {\n",
191
+ " \"alpha\": 0.3,\n",
192
+ " \"class_weight\": \"balanced\",\n",
193
+ " \"head\": \"l2+RidgeClassifier\"\n",
194
+ " }\n",
195
+ " },\n",
196
+ " {\n",
197
+ " \"dataset_key\": \"sider\",\n",
198
+ " \"task\": \"cardiac disorders\",\n",
199
+ " \"preset\": \"l2_ridge_a0.3_cwbalanced\",\n",
200
+ " \"preset_params\": {\n",
201
+ " \"alpha\": 0.3,\n",
202
+ " \"class_weight\": \"balanced\",\n",
203
+ " \"head\": \"l2+RidgeClassifier\"\n",
204
+ " }\n",
205
+ " },\n",
206
+ " {\n",
207
+ " \"dataset_key\": \"sider\",\n",
208
+ " \"task\": \"congenital, familial and genetic disorders\",\n",
209
+ " \"preset\": \"svd128_ridge_a0.001_cwbalanced\",\n",
210
+ " \"preset_params\": {\n",
211
+ " \"alpha\": 0.001,\n",
212
+ " \"class_weight\": \"balanced\",\n",
213
+ " \"head\": \"svd+RidgeClassifier\",\n",
214
+ " \"n_components\": 128\n",
215
+ " }\n",
216
+ " },\n",
217
+ " {\n",
218
+ " \"dataset_key\": \"sider\",\n",
219
+ " \"task\": \"ear and labyrinth disorders\",\n",
220
+ " \"preset\": \"l2_ridge_a0.3_cwbalanced\",\n",
221
+ " \"preset_params\": {\n",
222
+ " \"alpha\": 0.3,\n",
223
+ " \"class_weight\": \"balanced\",\n",
224
+ " \"head\": \"l2+RidgeClassifier\"\n",
225
+ " }\n",
226
+ " },\n",
227
+ " {\n",
228
+ " \"dataset_key\": \"sider\",\n",
229
+ " \"task\": \"endocrine disorders\",\n",
230
+ " \"preset\": \"minmax_ridge_a0.0001_cwnone\",\n",
231
+ " \"preset_params\": {\n",
232
+ " \"alpha\": 0.0001,\n",
233
+ " \"class_weight\": null,\n",
234
+ " \"head\": \"minmax+RidgeClassifier\"\n",
235
+ " }\n",
236
+ " },\n",
237
+ " {\n",
238
+ " \"dataset_key\": \"sider\",\n",
239
+ " \"task\": \"eye disorders\",\n",
240
+ " \"preset\": \"l2_ridge_a0.03_cwbalanced\",\n",
241
+ " \"preset_params\": {\n",
242
+ " \"alpha\": 0.03,\n",
243
+ " \"class_weight\": \"balanced\",\n",
244
+ " \"head\": \"l2+RidgeClassifier\"\n",
245
+ " }\n",
246
+ " },\n",
247
+ " {\n",
248
+ " \"dataset_key\": \"sider\",\n",
249
+ " \"task\": \"gastrointestinal disorders\",\n",
250
+ " \"preset\": \"l2_ridge_a3_cwnone\",\n",
251
+ " \"preset_params\": {\n",
252
+ " \"alpha\": 3,\n",
253
+ " \"class_weight\": null,\n",
254
+ " \"head\": \"l2+RidgeClassifier\"\n",
255
+ " }\n",
256
+ " },\n",
257
+ " {\n",
258
+ " \"dataset_key\": \"sider\",\n",
259
+ " \"task\": \"general disorders and administration site conditions\",\n",
260
+ " \"preset\": \"l2_ridge_a0.03_cwnone\",\n",
261
+ " \"preset_params\": {\n",
262
+ " \"alpha\": 0.03,\n",
263
+ " \"class_weight\": null,\n",
264
+ " \"head\": \"l2+RidgeClassifier\"\n",
265
+ " }\n",
266
+ " },\n",
267
+ " {\n",
268
+ " \"dataset_key\": \"sider\",\n",
269
+ " \"task\": \"hepatobiliary disorders\",\n",
270
+ " \"preset\": \"l2_ridge_a1_cwnone\",\n",
271
+ " \"preset_params\": {\n",
272
+ " \"alpha\": 1,\n",
273
+ " \"class_weight\": null,\n",
274
+ " \"head\": \"l2+RidgeClassifier\"\n",
275
+ " }\n",
276
+ " },\n",
277
+ " {\n",
278
+ " \"dataset_key\": \"sider\",\n",
279
+ " \"task\": \"immune system disorders\",\n",
280
+ " \"preset\": \"l2_ridge_a1_cwbalanced\",\n",
281
+ " \"preset_params\": {\n",
282
+ " \"alpha\": 1,\n",
283
+ " \"class_weight\": \"balanced\",\n",
284
+ " \"head\": \"l2+RidgeClassifier\"\n",
285
+ " }\n",
286
+ " },\n",
287
+ " {\n",
288
+ " \"dataset_key\": \"sider\",\n",
289
+ " \"task\": \"infections and infestations\",\n",
290
+ " \"preset\": \"l2_ridge_a3_cwnone\",\n",
291
+ " \"preset_params\": {\n",
292
+ " \"alpha\": 3,\n",
293
+ " \"class_weight\": null,\n",
294
+ " \"head\": \"l2+RidgeClassifier\"\n",
295
+ " }\n",
296
+ " },\n",
297
+ " {\n",
298
+ " \"dataset_key\": \"sider\",\n",
299
+ " \"task\": \"injury, poisoning and procedural complications\",\n",
300
+ " \"preset\": \"l2_ridge_a3_cwnone\",\n",
301
+ " \"preset_params\": {\n",
302
+ " \"alpha\": 3,\n",
303
+ " \"class_weight\": null,\n",
304
+ " \"head\": \"l2+RidgeClassifier\"\n",
305
+ " }\n",
306
+ " },\n",
307
+ " {\n",
308
+ " \"dataset_key\": \"sider\",\n",
309
+ " \"task\": \"investigations\",\n",
310
+ " \"preset\": \"std_l2_ridge_a0.1_cwbalanced\",\n",
311
+ " \"preset_params\": {\n",
312
+ " \"alpha\": 0.1,\n",
313
+ " \"class_weight\": \"balanced\",\n",
314
+ " \"head\": \"std_l2+RidgeClassifier\"\n",
315
+ " }\n",
316
+ " },\n",
317
+ " {\n",
318
+ " \"dataset_key\": \"sider\",\n",
319
+ " \"task\": \"metabolism and nutrition disorders\",\n",
320
+ " \"preset\": \"l2_ridge_a3_cwbalanced\",\n",
321
+ " \"preset_params\": {\n",
322
+ " \"alpha\": 3,\n",
323
+ " \"class_weight\": \"balanced\",\n",
324
+ " \"head\": \"l2+RidgeClassifier\"\n",
325
+ " }\n",
326
+ " },\n",
327
+ " {\n",
328
+ " \"dataset_key\": \"sider\",\n",
329
+ " \"task\": \"musculoskeletal and connective tissue disorders\",\n",
330
+ " \"preset\": \"l2_ridge_a0.03_cwnone\",\n",
331
+ " \"preset_params\": {\n",
332
+ " \"alpha\": 0.03,\n",
333
+ " \"class_weight\": null,\n",
334
+ " \"head\": \"l2+RidgeClassifier\"\n",
335
+ " }\n",
336
+ " },\n",
337
+ " {\n",
338
+ " \"dataset_key\": \"sider\",\n",
339
+ " \"task\": \"neoplasms benign, malignant and unspecified (incl cysts and polyps)\",\n",
340
+ " \"preset\": \"l2_ridge_a1e-05_cwbalanced\",\n",
341
+ " \"preset_params\": {\n",
342
+ " \"alpha\": 1e-05,\n",
343
+ " \"class_weight\": \"balanced\",\n",
344
+ " \"head\": \"l2+RidgeClassifier\"\n",
345
+ " }\n",
346
+ " },\n",
347
+ " {\n",
348
+ " \"dataset_key\": \"sider\",\n",
349
+ " \"task\": \"nervous system disorders\",\n",
350
+ " \"preset\": \"minmax_ridge_a30_cwbalanced\",\n",
351
+ " \"preset_params\": {\n",
352
+ " \"alpha\": 30,\n",
353
+ " \"class_weight\": \"balanced\",\n",
354
+ " \"head\": \"minmax+RidgeClassifier\"\n",
355
+ " }\n",
356
+ " },\n",
357
+ " {\n",
358
+ " \"dataset_key\": \"sider\",\n",
359
+ " \"task\": \"pregnancy, puerperium and perinatal conditions\",\n",
360
+ " \"preset\": \"svd32_ridge_a0.001_cwnone\",\n",
361
+ " \"preset_params\": {\n",
362
+ " \"alpha\": 0.001,\n",
363
+ " \"class_weight\": null,\n",
364
+ " \"head\": \"svd+RidgeClassifier\",\n",
365
+ " \"n_components\": 32\n",
366
+ " }\n",
367
+ " },\n",
368
+ " {\n",
369
+ " \"dataset_key\": \"sider\",\n",
370
+ " \"task\": \"product issues\",\n",
371
+ " \"preset\": \"l2_ridge_a1000_cwbalanced\",\n",
372
+ " \"preset_params\": {\n",
373
+ " \"alpha\": 1000,\n",
374
+ " \"class_weight\": \"balanced\",\n",
375
+ " \"head\": \"l2+RidgeClassifier\"\n",
376
+ " }\n",
377
+ " },\n",
378
+ " {\n",
379
+ " \"dataset_key\": \"sider\",\n",
380
+ " \"task\": \"psychiatric disorders\",\n",
381
+ " \"preset\": \"l2_ridge_a1_cwbalanced\",\n",
382
+ " \"preset_params\": {\n",
383
+ " \"alpha\": 1,\n",
384
+ " \"class_weight\": \"balanced\",\n",
385
+ " \"head\": \"l2+RidgeClassifier\"\n",
386
+ " }\n",
387
+ " },\n",
388
+ " {\n",
389
+ " \"dataset_key\": \"sider\",\n",
390
+ " \"task\": \"renal and urinary disorders\",\n",
391
+ " \"preset\": \"l2_ridge_a0.1_cwnone\",\n",
392
+ " \"preset_params\": {\n",
393
+ " \"alpha\": 0.1,\n",
394
+ " \"class_weight\": null,\n",
395
+ " \"head\": \"l2+RidgeClassifier\"\n",
396
+ " }\n",
397
+ " },\n",
398
+ " {\n",
399
+ " \"dataset_key\": \"sider\",\n",
400
+ " \"task\": \"reproductive system and breast disorders\",\n",
401
+ " \"preset\": \"l2_ridge_a3e-05_cwbalanced\",\n",
402
+ " \"preset_params\": {\n",
403
+ " \"alpha\": 3e-05,\n",
404
+ " \"class_weight\": \"balanced\",\n",
405
+ " \"head\": \"l2+RidgeClassifier\"\n",
406
+ " }\n",
407
+ " },\n",
408
+ " {\n",
409
+ " \"dataset_key\": \"sider\",\n",
410
+ " \"task\": \"respiratory, thoracic and mediastinal disorders\",\n",
411
+ " \"preset\": \"l2_ridge_a1_cwnone\",\n",
412
+ " \"preset_params\": {\n",
413
+ " \"alpha\": 1,\n",
414
+ " \"class_weight\": null,\n",
415
+ " \"head\": \"l2+RidgeClassifier\"\n",
416
+ " }\n",
417
+ " },\n",
418
+ " {\n",
419
+ " \"dataset_key\": \"sider\",\n",
420
+ " \"task\": \"skin and subcutaneous tissue disorders\",\n",
421
+ " \"preset\": \"l2_ridge_a1000_cwbalanced\",\n",
422
+ " \"preset_params\": {\n",
423
+ " \"alpha\": 1000,\n",
424
+ " \"class_weight\": \"balanced\",\n",
425
+ " \"head\": \"l2+RidgeClassifier\"\n",
426
+ " }\n",
427
+ " },\n",
428
+ " {\n",
429
+ " \"dataset_key\": \"sider\",\n",
430
+ " \"task\": \"social circumstances\",\n",
431
+ " \"preset\": \"l2_ridge_a3_cwbalanced\",\n",
432
+ " \"preset_params\": {\n",
433
+ " \"alpha\": 3,\n",
434
+ " \"class_weight\": \"balanced\",\n",
435
+ " \"head\": \"l2+RidgeClassifier\"\n",
436
+ " }\n",
437
+ " },\n",
438
+ " {\n",
439
+ " \"dataset_key\": \"sider\",\n",
440
+ " \"task\": \"surgical and medical procedures\",\n",
441
+ " \"preset\": \"std_l2_ridge_a1e-05_cwnone\",\n",
442
+ " \"preset_params\": {\n",
443
+ " \"alpha\": 1e-05,\n",
444
+ " \"class_weight\": null,\n",
445
+ " \"head\": \"std_l2+RidgeClassifier\"\n",
446
+ " }\n",
447
+ " },\n",
448
+ " {\n",
449
+ " \"dataset_key\": \"sider\",\n",
450
+ " \"task\": \"vascular disorders\",\n",
451
+ " \"preset\": \"l2_ridge_a0.0003_cwnone\",\n",
452
+ " \"preset_params\": {\n",
453
+ " \"alpha\": 0.0003,\n",
454
+ " \"class_weight\": null,\n",
455
+ " \"head\": \"l2+RidgeClassifier\"\n",
456
+ " }\n",
457
+ " }\n",
458
+ "]\n",
459
+ "\n",
460
+ "print(json.dumps(TASK_SETTINGS, indent=2))\n",
461
+ "print(f'Loaded {len(PHYSIOLOGICAL_TASK_PRESETS)} Tox21/SIDER task-specific settings')\n"
462
+ ]
463
+ },
464
+ {
465
+ "cell_type": "code",
466
+ "execution_count": null,
467
+ "metadata": {},
468
+ "outputs": [],
469
+ "source": [
470
+ "def encode_rmmol_smiles(smiles):\n",
471
+ " if not RMMOL_CKPT:\n",
472
+ " raise RuntimeError('Set RMMOL_CKPT to the downloaded RMMol checkpoint path before encoding.')\n",
473
+ " sys.path.insert(0, str(RMMOL_TRUE_LOADER_DIR))\n",
474
+ " from rmmol_true_loader import encode_smiles_true_loader\n",
475
+ " embeddings, valid_indices, cfg = encode_smiles_true_loader(\n",
476
+ " list(smiles), checkpoint=Path(RMMOL_CKPT), batch_size=BATCH_SIZE, device=DEVICE\n",
477
+ " )\n",
478
+ " return embeddings, np.asarray(valid_indices, dtype=int), cfg\n",
479
+ "\n",
480
+ "def read_splits(domain, dataset, prefix):\n",
481
+ " frames = []\n",
482
+ " for split in ['train', 'val', 'test']:\n",
483
+ " frame = pd.read_csv(RELEASE_DIR / 'dataset' / domain / dataset / f'{prefix}_{split}.csv')\n",
484
+ " frame['_split'] = split\n",
485
+ " frames.append(frame)\n",
486
+ " return pd.concat(frames, ignore_index=True)\n",
487
+ "\n",
488
+ "def fit_auc(head_factory, x, y, folds):\n",
489
+ " scores = []\n",
490
+ " for train_idx, test_idx in folds:\n",
491
+ " model = head_factory()\n",
492
+ " model.fit(x[train_idx], y[train_idx])\n",
493
+ " pred = model.decision_function(x[test_idx])\n",
494
+ " scores.append(roc_auc_score(y[test_idx], pred))\n",
495
+ " return float(np.mean(scores)), float(np.std(scores, ddof=0))\n"
496
+ ]
497
+ },
498
+ {
499
+ "cell_type": "markdown",
500
+ "metadata": {},
501
+ "source": [
502
+ "## Biophysics examples"
503
+ ]
504
+ },
505
+ {
506
+ "cell_type": "code",
507
+ "execution_count": null,
508
+ "metadata": {},
509
+ "outputs": [],
510
+ "source": [
511
+ "bbbp = pd.read_csv(RELEASE_DIR / 'dataset' / 'biophysics' / 'bbbp' / 'bbbp.csv')\n",
512
+ "x_bbbp, valid_bbbp, _ = encode_rmmol_smiles(bbbp['smiles'])\n",
513
+ "y_bbbp = bbbp.iloc[valid_bbbp]['p_np'].astype(int).to_numpy()\n",
514
+ "folds = list(KFold(n_splits=5, shuffle=True, random_state=42).split(x_bbbp))\n",
515
+ "bbbp_auc, bbbp_std = fit_auc(\n",
516
+ " lambda: make_pipeline(StandardScaler(), RidgeClassifier(alpha=1.0, class_weight=None, solver='lsqr')),\n",
517
+ " x_bbbp, y_bbbp, folds\n",
518
+ ")\n",
519
+ "print({'dataset': 'bbbp', 'metric': 'ROC-AUC', 'mean': bbbp_auc, 'std': bbbp_std, 'n': len(y_bbbp)})\n",
520
+ "\n",
521
+ "hiv = read_splits('biophysics', 'hiv', 'hiv')\n",
522
+ "x_hiv, valid_hiv, _ = encode_rmmol_smiles(hiv['smiles'])\n",
523
+ "hiv_valid = hiv.iloc[valid_hiv].reset_index(drop=True)\n",
524
+ "y_hiv = hiv_valid['HIV_active'].astype(int).to_numpy()\n",
525
+ "trainval = hiv_valid['_split'].isin(['train', 'val']).to_numpy()\n",
526
+ "test = hiv_valid['_split'].eq('test').to_numpy()\n",
527
+ "hiv_head = make_pipeline(Normalizer(), RidgeClassifier(alpha=1.6788e-05, class_weight='balanced', solver='lsqr'))\n",
528
+ "hiv_head.fit(x_hiv[trainval], y_hiv[trainval])\n",
529
+ "hiv_auc = roc_auc_score(y_hiv[test], hiv_head.decision_function(x_hiv[test]))\n",
530
+ "print({'dataset': 'hiv', 'metric': 'ROC-AUC', 'test': float(hiv_auc), 'n_test': int(test.sum())})\n"
531
+ ]
532
+ },
533
+ {
534
+ "cell_type": "markdown",
535
+ "metadata": {},
536
+ "source": [
537
+ "## Physical chemistry example"
538
+ ]
539
+ },
540
+ {
541
+ "cell_type": "code",
542
+ "execution_count": null,
543
+ "metadata": {},
544
+ "outputs": [],
545
+ "source": [
546
+ "def regression_cv(domain, dataset, prefix, label, alpha):\n",
547
+ " frame = read_splits(domain, dataset, prefix)\n",
548
+ " x, valid, _ = encode_rmmol_smiles(frame['smiles'])\n",
549
+ " y = frame.iloc[valid][label].astype(float).to_numpy()\n",
550
+ " folds = KFold(n_splits=5, shuffle=True, random_state=42).split(x)\n",
551
+ " rmses = []\n",
552
+ " for train_idx, test_idx in folds:\n",
553
+ " model = make_pipeline(StandardScaler(), Ridge(alpha=alpha))\n",
554
+ " model.fit(x[train_idx], y[train_idx])\n",
555
+ " pred = model.predict(x[test_idx])\n",
556
+ " rmses.append(mean_squared_error(y[test_idx], pred, squared=False))\n",
557
+ " return float(np.mean(rmses)), float(np.std(rmses, ddof=0)), len(y)\n",
558
+ "\n",
559
+ "esol_rmse, esol_std, esol_n = regression_cv(\n",
560
+ " 'physical_chemistry', 'esol', 'esol', 'ESOL predicted log solubility in mols per litre', 0.003\n",
561
+ ")\n",
562
+ "print({'dataset': 'esol', 'metric': 'RMSE', 'mean': esol_rmse, 'std': esol_std, 'n': esol_n})\n"
563
+ ]
564
+ },
565
+ {
566
+ "cell_type": "markdown",
567
+ "metadata": {},
568
+ "source": [
569
+ "## Quantum mechanics example"
570
+ ]
571
+ },
572
+ {
573
+ "cell_type": "code",
574
+ "execution_count": null,
575
+ "metadata": {},
576
+ "outputs": [],
577
+ "source": [
578
+ "qm9_tasks = ['mu', 'alpha', 'homo', 'lumo', 'gap', 'r2', 'zpve', 'u0', 'u298', 'h298', 'g298', 'cv']\n",
579
+ "qm9 = read_splits('quantum_mechanics', 'qm9', 'qm9')\n",
580
+ "x_qm9, valid_qm9, _ = encode_rmmol_smiles(qm9['smiles'])\n",
581
+ "y_qm9 = qm9.iloc[valid_qm9][qm9_tasks].astype(float).to_numpy()\n",
582
+ "folds = KFold(n_splits=5, shuffle=True, random_state=42).split(x_qm9)\n",
583
+ "maes = []\n",
584
+ "for train_idx, test_idx in folds:\n",
585
+ " model = make_pipeline(StandardScaler(), Ridge(alpha=0.01))\n",
586
+ " model.fit(x_qm9[train_idx], y_qm9[train_idx])\n",
587
+ " pred = model.predict(x_qm9[test_idx])\n",
588
+ " maes.append(mean_absolute_error(y_qm9[test_idx], pred))\n",
589
+ "print({'dataset': 'qm9', 'metric': 'MAE', 'mean': float(np.mean(maes)), 'std': float(np.std(maes, ddof=0)), 'n': len(y_qm9)})\n"
590
+ ]
591
+ }
592
+ ],
593
+ "metadata": {
594
+ "kernelspec": {
595
+ "display_name": "Python 3",
596
+ "language": "python",
597
+ "name": "python3"
598
+ },
599
+ "language_info": {
600
+ "name": "python",
601
+ "pygments_lexer": "ipython3"
602
+ }
603
+ },
604
+ "nbformat": 4,
605
+ "nbformat_minor": 5
606
+ }