Newer
Older
Tim O'Donnell
committed
"execution_count": 1,
Tim O'Donnell
committed
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"Using Theano backend.\n",
"/Users/tim/miniconda3/envs/py3k/lib/python3.5/site-packages/matplotlib/__init__.py:872: UserWarning: axes.color_cycle is deprecated and replaced with axes.prop_cycle; please use the latter.\n",
" warnings.warn(self.msg_depr % (key, alt_key))\n",
"/Users/tim/miniconda3/envs/py3k/lib/python3.5/site-packages/sklearn/cross_validation.py:44: DeprecationWarning: This module was deprecated in version 0.18 in favor of the model_selection module into which all the refactored classes and functions are moved. Also note that the interface of the new CV iterators are different from that of this module. This module will be removed in 0.20.\n",
" \"This module will be removed in 0.20.\", DeprecationWarning)\n"
]
}
],
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
"source": [
"import mhcflurry\n",
"import numpy\n",
"import seaborn\n",
"import logging\n",
"from matplotlib import pyplot\n",
"\n",
"% matplotlib inline\n",
"logging.basicConfig(level=\"DEBUG\")\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Making predictions\n",
"Note: if you haven't already, run `mhcflurry-downloads fetch` in a shell to download the trained models."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Simplest way to run predictions: `mhcflurry.predict()`"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {
"collapsed": false
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Help on function predict in module mhcflurry.predict:\n",
"\n",
Tim O'Donnell
committed
"predict(alleles, peptides, predictor=None)\n",
" Make predictions across all combinations of the specified alleles and\n",
" peptides.\n",
" \n",
" Parameters\n",
" ----------\n",
" alleles : list of str\n",
" Names of alleles to make predictions for.\n",
" \n",
" peptides : list of str\n",
" Peptide amino acid sequences.\n",
" \n",
Tim O'Donnell
committed
" predictor : Predictor to use. Defaults to downloaded Class1SingleModelMultiAllelePredictor.\n",
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
" \n",
" Returns DataFrame with columns \"Allele\", \"Peptide\", and \"Prediction\"\n",
"\n"
]
}
],
"source": [
"help(mhcflurry.predict)"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>Allele</th>\n",
" <th>Peptide</th>\n",
" <th>Prediction</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>HLA-A0201</td>\n",
" <td>SIINFEKL</td>\n",
" <td>10672.347656</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>HLA-A0201</td>\n",
" <td>SIINFEQL</td>\n",
" <td>7828.974121</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"</div>"
],
"text/plain": [
" Allele Peptide Prediction\n",
"0 HLA-A0201 SIINFEKL 10672.347656\n",
"1 HLA-A0201 SIINFEQL 7828.974121"
]
},
"execution_count": 3,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"mhcflurry.predict(alleles=[\"HLA-A0201\"], peptides=[\"SIINFEKL\", \"SIINFEQL\"])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
" "
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Instantiating a model"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"array([ 10672.34765625, 30577.02539062, 10565.78222656], dtype=float32)"
]
},
"execution_count": 4,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
Tim O'Donnell
committed
"model = mhcflurry.class1_allele_specific.class1_single_model_multi_allele_predictor.from_allele_name(\"HLA-A0201\")\n",
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
"model.predict([\"SIINFEKL\", \"SIQNPEKP\", \"SYNFPEPI\"])\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
" "
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Instantiating a model from a custom set of models on disk"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
Tim O'Donnell
committed
"'/Users/tim/Library/Application Support/mhcflurry/4/0.1.0/models_class1_allele_specific_single/'"
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
]
},
"execution_count": 5,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"models_dir = mhcflurry.downloads.get_path(\"models_class1_allele_specific_single\")\n",
"models_dir"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"array([ 10672.34765625, 30577.02539062, 10565.78222656], dtype=float32)"
]
},
"execution_count": 6,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# Make a Loader first\n",
Tim O'Donnell
committed
"loader = mhcflurry.Class1SingleModelMultiAllelePredictor.load_from_download_directory(\n",
" models_dir)\n",
"model = loader.predictor_for_allele(\"HLA-A0201\")\n",
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
"model.predict([\"SIINFEKL\", \"SIQNPEKP\", \"SYNFPEPI\"])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
" "
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Loading a `Dataset`"
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
Tim O'Donnell
committed
"AffinityMeasurementDataset(n=192550, alleles=['ELA-A1', 'Gogo-B0101', 'H-2-DB', 'H-2-DD', 'H-2-KB', 'H-2-KBM8', 'H-2-KD', 'H-2-KK', 'H-2-LD', 'H-2-LQ', 'HLA-A0101', 'HLA-A0201', 'HLA-A0202', 'HLA-A0203', 'HLA-A0204', 'HLA-A0205', 'HLA-A0206', 'HLA-A0207', 'HLA-A0210', 'HLA-A0211', 'HLA-A0212', 'HLA-A0216', 'HLA-A0217', 'HLA-A0219', 'HLA-A0250', 'HLA-A0301', 'HLA-A0302', 'HLA-A0319', 'HLA-A1', 'HLA-A11', 'HLA-A1101', 'HLA-A1102', 'HLA-A2', 'HLA-A2301', 'HLA-A24', 'HLA-A2402', 'HLA-A2403', 'HLA-A2501', 'HLA-A26', 'HLA-A2601', 'HLA-A2602', 'HLA-A2603', 'HLA-A2902', 'HLA-A3', 'HLA-A3/11', 'HLA-A3001', 'HLA-A3002', 'HLA-A3101', 'HLA-A3201', 'HLA-A3207', 'HLA-A3215', 'HLA-A3301', 'HLA-A6601', 'HLA-A6801', 'HLA-A6802', 'HLA-A6823', 'HLA-A6901', 'HLA-A7401', 'HLA-A8001', 'HLA-B0702', 'HLA-B0801', 'HLA-B0802', 'HLA-B0803', 'HLA-B1401', 'HLA-B1402', 'HLA-B1501', 'HLA-B1502', 'HLA-B1503', 'HLA-B1509', 'HLA-B1517', 'HLA-B1542', 'HLA-B1801', 'HLA-B27', 'HLA-B2701', 'HLA-B2702', 'HLA-B2703', 'HLA-B2704', 'HLA-B2705', 'HLA-B2706', 'HLA-B2710', 'HLA-B2720', 'HLA-B3501', 'HLA-B3503', 'HLA-B3508', 'HLA-B3701', 'HLA-B3801', 'HLA-B39', 'HLA-B3901', 'HLA-B40', 'HLA-B4001', 'HLA-B4002', 'HLA-B4013', 'HLA-B4201', 'HLA-B4202', 'HLA-B44', 'HLA-B4402', 'HLA-B4403', 'HLA-B4501', 'HLA-B4506', 'HLA-B4601', 'HLA-B4801', 'HLA-B51', 'HLA-B5101', 'HLA-B5201', 'HLA-B5301', 'HLA-B5401', 'HLA-B5701', 'HLA-B5702', 'HLA-B5703', 'HLA-B58', 'HLA-B5801', 'HLA-B5802', 'HLA-B60', 'HLA-B62', 'HLA-B7', 'HLA-B7301', 'HLA-B8', 'HLA-B8101', 'HLA-B8301', 'HLA-BOLA102101', 'HLA-BOLA200801', 'HLA-BOLA201201', 'HLA-BOLA402401', 'HLA-BOLA601301', 'HLA-BOLA601302', 'HLA-BOLAHD6', 'HLA-C0303', 'HLA-C0401', 'HLA-C0501', 'HLA-C0602', 'HLA-C0702', 'HLA-C0802', 'HLA-C1', 'HLA-C1203', 'HLA-C1402', 'HLA-C1502', 'HLA-C4', 'HLA-E0101', 'HLA-E0103', 'HLA-EQCA100101', 'HLA-RT1A', 'HLA-RT1BL', 'HLA-SLA10401', 'Mamu-A01', 'Mamu-A02', 'Mamu-A07', 'Mamu-A100101', 'Mamu-A100201', 'Mamu-A101101', 'Mamu-A11', 'Mamu-A20102', 'Mamu-A2201', 'Mamu-A2601', 'Mamu-A70103', 'Mamu-B01', 'Mamu-B01704', 'Mamu-B03', 'Mamu-B04', 'Mamu-B06502', 'Mamu-B08', 'Mamu-B1001', 'Mamu-B17', 'Mamu-B3901', 'Mamu-B52', 'Mamu-B6601', 'Mamu-B8301', 'Mamu-B8701', 'Patr-A0101', 'Patr-A0301', 'Patr-A0401', 'Patr-A0602', 'Patr-A0701', 'Patr-A0901', 'Patr-B0101', 'Patr-B0901', 'Patr-B1301', 'Patr-B1701', 'Patr-B2401'])"
]
},
"execution_count": 7,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
Tim O'Donnell
committed
"full_training_data = mhcflurry.AffinityMeasurementDataset.from_csv(\n",
" mhcflurry.downloads.get_path(\"data_combined_iedb_kim2014\", \"combined_human_class1_dataset.csv\"))\n",
"full_training_data"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"(179692, 137654, 27680)"
]
},
"execution_count": 8,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
Tim O'Donnell
committed
"kim2014_full = mhcflurry.AffinityMeasurementDataset.from_csv(\n",
" mhcflurry.downloads.get_path(\"data_kim2014\", \"bdata.20130222.mhci.public.1.txt\"))\n",
"\n",
Tim O'Donnell
committed
"kim2014_train = mhcflurry.AffinityMeasurementDataset.from_csv(\n",
" mhcflurry.downloads.get_path(\"data_kim2014\", \"bdata.2009.mhci.public.1.txt\"))\n",
Tim O'Donnell
committed
"kim2014_test = mhcflurry.AffinityMeasurementDataset.from_csv(\n",
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
" mhcflurry.downloads.get_path(\"data_kim2014\", \"bdata.2013.mhci.public.blind.1.txt\"))\n",
"\n",
"len(kim2014_full), len(kim2014_train), len(kim2014_test)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
" "
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Predicting affinities from a `Dataset`\n"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"array([ 3514.14550781, 12429.5390625 , 4227.02197266, ...,\n",
" 5949.32763672, 17837.0859375 , 6724.96728516], dtype=float32)"
]
},
"execution_count": 9,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
Tim O'Donnell
committed
"model = loader.predictor_for_allele(\"HLA-A0201\")\n",
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
"model.predict(kim2014_train.get_allele(\"HLA-A0201\").peptides)\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
" "
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Fit a model"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {
"collapsed": false
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Help on class Class1BindingPredictor in module mhcflurry.class1_allele_specific.class1_binding_predictor:\n",
"\n",
"class Class1BindingPredictor(mhcflurry.class1_allele_specific.class1_allele_specific_kmer_ic50_predictor_base.Class1AlleleSpecificKmerIC50PredictorBase)\n",
" | Allele-specific Class I MHC binding predictor which uses\n",
" | fixed-length (k-mer) index encoding for inputs and outputs\n",
" | a value between 0 and 1 (where 1 is the strongest binder).\n",
" | \n",
" | Method resolution order:\n",
" | Class1BindingPredictor\n",
" | mhcflurry.class1_allele_specific.class1_allele_specific_kmer_ic50_predictor_base.Class1AlleleSpecificKmerIC50PredictorBase\n",
" | mhcflurry.ic50_predictor_base.IC50PredictorBase\n",
" | builtins.object\n",
" | \n",
" | Methods defined here:\n",
" | \n",
" | __getstate__(self)\n",
" | \n",
" | __init__(self, model=None, name=None, max_ic50=50000.0, allow_unknown_amino_acids=True, kmer_size=9, n_amino_acids=20, verbose=False, **hyperparameters)\n",
" | Initialize self. See help(type(self)) for accurate signature.\n",
" | \n",
" | __setstate__(self, state)\n",
" | \n",
" | fit_kmer_encoded_arrays(self, X, ic50, sample_weights=None, right_censoring_mask=None, X_pretrain=None, ic50_pretrain=None, sample_weights_pretrain=None, n_random_negative_samples=None, pretrain_decay=None, n_training_epochs=None, batch_size=None, verbose=False)\n",
" | Train predictive model from index encoding of fixed length k-mer\n",
" | peptides.\n",
" | \n",
" | Parameters\n",
" | ----------\n",
" | X : array\n",
" | Training data with shape (n_samples, n_dims)\n",
" | \n",
" | ic50 : array\n",
" | Training IC50 values with shape (n_samples,)\n",
" | \n",
" | sample_weights : array\n",
" | Weight of each training sample with shape (n_samples,)\n",
" | \n",
" | right_censoring_mask : array, optional\n",
" | Boolean array which indicates whether each IC50 value is actually\n",
" | right censored (a lower bound on the true value). Censored values\n",
" | are transformed during training by sampling between the observed\n",
" | and maximum values on each iteration.\n",
" | \n",
" | X_pretrain : array\n",
" | Extra samples used for soft pretraining of the predictor,\n",
" | should have same number of dimensions as X.\n",
" | During training the weights of these samples will decay after\n",
" | each epoch.\n",
" | \n",
" | ic50_pretrain : array\n",
" | IC50 values for extra samples, shape\n",
" | \n",
" | pretrain_decay : int -> float function\n",
" | decay function for pretraining, mapping epoch number to decay\n",
" | factor\n",
" | \n",
" | sample_weights_pretrain : array\n",
" | Initial weights for the rows of X_pretrain. If not specified then\n",
" | initialized to ones.\n",
" | \n",
" | n_random_negative_samples : int\n",
" | Number of random samples to generate as negative examples.\n",
" | \n",
" | n_training_epochs : int\n",
" | \n",
" | verbose : bool\n",
" | \n",
" | batch_size : int\n",
" | \n",
" | get_weights(self)\n",
" | Returns weights, which can be passed to set_weights later.\n",
" | \n",
" | predict_ic50_for_kmer_encoded_array(self, X)\n",
" | Given an encoded array of amino acid indices,\n",
" | returns a vector of IC50 predictions.\n",
" | \n",
" | predict_scores_for_kmer_encoded_array(self, X)\n",
" | Given an encoded array of amino acid indices, returns a vector\n",
" | of affinity scores (values between 0 and 1).\n",
" | \n",
" | set_weights(self, weights)\n",
" | Reset the model weights.\n",
" | \n",
" | ----------------------------------------------------------------------\n",
" | Data and other attributes defined here:\n",
" | \n",
" | fit_hyperparameter_defaults = <mhcflurry.hyperparameters.Hyperparamete...\n",
" | \n",
" | hyperparameter_defaults = <mhcflurry.hyperparameters.HyperparameterDef...\n",
" | \n",
" | network_hyperparameter_defaults = <mhcflurry.hyperparameters.Hyperpara...\n",
" | \n",
" | ----------------------------------------------------------------------\n",
" | Methods inherited from mhcflurry.class1_allele_specific.class1_allele_specific_kmer_ic50_predictor_base.Class1AlleleSpecificKmerIC50PredictorBase:\n",
" | \n",
" | __repr__(self)\n",
" | Return repr(self).\n",
" | \n",
" | __str__(self)\n",
" | Return str(self).\n",
" | \n",
" | encode_peptides(self, peptides)\n",
" | \n",
" | fit_dataset(self, dataset, pretraining_dataset=None, sample_censored_affinities=False, **kwargs)\n",
" | Fit the model parameters on the given training data.\n",
" | \n",
" | Parameters\n",
" | ----------\n",
Tim O'Donnell
committed
" | dataset : AffinityMeasurementDataset\n",
Tim O'Donnell
committed
" | pretraining_dataset : AffinityMeasurementDataset\n",
" | \n",
" | sample_censored_affinities : bool\n",
Tim O'Donnell
committed
" | If a column named 'inequality' is in the AffinityMeasurementDataset then every\n",
" | peptide with a value of '>' on each training epoch, gets a\n",
" | randomly sampled IC50 between its observed value and the\n",
" | max_ic50 of the predictor. Default is False.\n",
" | \n",
" | **kwargs : dict\n",
" | Extra arguments are passed on to the fit_encoded_kmer_arrays()\n",
" | method.\n",
" | \n",
Tim O'Donnell
committed
" | predict_scores(self, peptides)\n",
" | Given a list of peptides of any length, returns an array of predicted\n",
" | normalized affinity values. Unlike IC50, a higher value here\n",
" | means a stronger affinity. Peptides of lengths other than 9 are\n",
" | transformed into a set of k-mers either by deleting or inserting\n",
Tim O'Donnell
committed
" | amino acid characters. The prediction for a single peptides will be\n",
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
" | the average of expanded k-mers.\n",
" | \n",
" | ----------------------------------------------------------------------\n",
" | Data descriptors inherited from mhcflurry.class1_allele_specific.class1_allele_specific_kmer_ic50_predictor_base.Class1AlleleSpecificKmerIC50PredictorBase:\n",
" | \n",
" | amino_acids\n",
" | Amino acid alphabet used for encoding peptides, may include\n",
" | \"X\" if allow_unknown_amino_acids is True.\n",
" | \n",
" | max_amino_acid_encoding_value\n",
" | \n",
" | ----------------------------------------------------------------------\n",
" | Methods inherited from mhcflurry.ic50_predictor_base.IC50PredictorBase:\n",
" | \n",
" | fit_dictionary(self, peptide_to_ic50_dict, **kwargs)\n",
" | Fit the model parameters using the given peptide->IC50 dictionary,\n",
" | all samples are given the same weight.\n",
" | \n",
" | Parameters\n",
" | ----------\n",
" | peptide_to_ic50_dict : dict\n",
" | Dictionary that maps peptides to IC50 values.\n",
" | \n",
" | fit_sequences(self, peptides, affinities, sample_weights=None, alleles=None, **kwargs)\n",
" | \n",
" | predict(self, peptides)\n",
" | Predict IC50 affinities for peptides of any length\n",
" | \n",
" | ----------------------------------------------------------------------\n",
" | Data descriptors inherited from mhcflurry.ic50_predictor_base.IC50PredictorBase:\n",
" | \n",
" | __dict__\n",
" | dictionary for instance variables (if defined)\n",
" | \n",
" | __weakref__\n",
" | list of weak references to the object (if defined)\n",
"\n"
]
}
],
"source": [
"help(mhcflurry.class1_allele_specific.Class1BindingPredictor)"
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
Tim O'Donnell
committed
"AffinityMeasurementDataset(n=3040, alleles=['HLA-A3301'])"
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
]
},
"execution_count": 11,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"train_data = kim2014_train.get_allele(\"HLA-A3301\")\n",
"train_data"
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"{'activation': 'tanh',\n",
" 'batch_normalization': True,\n",
" 'batch_size': 128,\n",
" 'dropout_probability': 0.0,\n",
" 'embedding_output_dim': 32,\n",
" 'fraction_negative': 0.0,\n",
" 'init': 'glorot_uniform',\n",
" 'kmer_size': 9,\n",
" 'layer_sizes': [64],\n",
" 'loss': 'mse',\n",
" 'max_ic50': 50000.0,\n",
" 'n_training_epochs': 250,\n",
" 'optimizer': 'rmsprop',\n",
" 'output_activation': 'sigmoid',\n",
" 'pretrain_decay': 'numpy.exp(-epoch)'}"
]
},
"execution_count": 12,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# We'll use the default hyper parameters here. Could also specify them as kwargs:\n",
"new_model = mhcflurry.class1_allele_specific.Class1BindingPredictor()\n",
"new_model.hyperparameters"
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {
"collapsed": false
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
Tim O'Donnell
committed
"CPU times: user 1min 11s, sys: 5.66 s, total: 1min 16s\n",
"Wall time: 1min 12s\n"
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
]
}
],
"source": [
"# This will run faster if you have a GPU.\n",
"%time new_model.fit_dataset(train_data)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
" "
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Evaluate the fit model on held-out test data"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Generate predictions"
]
},
{
"cell_type": "code",
"execution_count": 14,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
Tim O'Donnell
committed
"<matplotlib.text.Text at 0x121829dd8>"
]
},
"execution_count": 14,
"metadata": {},
"output_type": "execute_result"
},
{
"data": {
Tim O'Donnell
committed
"image/png": "iVBORw0KGgoAAAANSUhEUgAAAesAAAFtCAYAAAAnGkJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nBieumzNQAAAAAElFTkSuQmCC\n",
Tim O'Donnell
committed
"<matplotlib.figure.Figure at 0x11d2d7da0>"
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"test_data = kim2014_test.get_allele(\"HLA-A3301\")\n",
"predictions = new_model.predict(test_data.peptides)\n",
"\n",
"seaborn.set_context('notebook')\n",
"seaborn.regplot(numpy.log10(test_data.affinities), numpy.log10(predictions))\n",
"pyplot.xlim(xmin=0)\n",
"pyplot.ylim(ymin=0)\n",
"pyplot.xlabel(\"Measured affinity (log10 nM)\")\n",
"pyplot.ylabel(\"Predicted affinity (log10 nM)\")\n",
"pyplot.title(\"MHCflurry on test data\")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
" "
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Calculate AUC, F1, and Kendall's Tau scores"
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {
"collapsed": false
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Help on function make_scores in module mhcflurry.class1_allele_specific.scoring:\n",
"\n",
"make_scores(ic50_y, ic50_y_pred, sample_weight=None, threshold_nm=500, max_ic50=50000)\n",
" Calculate AUC, F1, and Kendall Tau scores.\n",
" \n",
" Parameters\n",
" -----------\n",
" ic50_y : float list\n",
" true IC50s (i.e. affinities)\n",
" \n",
" ic50_y_pred : float list\n",
" predicted IC50s\n",
" \n",
" sample_weight : float list [optional]\n",
" \n",
" threshold_nm : float [optional]\n",
" \n",
" max_ic50 : float [optional]\n",
" \n",
" Returns\n",
" -----------\n",
" dict with entries \"auc\", \"f1\", \"tau\"\n",
"\n"
]
}
],
"source": [
"help(mhcflurry.class1_allele_specific.scoring.make_scores)"
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
Tim O'Donnell
committed
"{'auc': 0.88567942942942945,\n",
" 'f1': 0.73814432989690715,\n",
" 'tau': 0.4912073299323329}"
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
]
},
"execution_count": 16,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"mhcflurry.class1_allele_specific.scoring.make_scores(test_data.affinities, predictions)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
" "
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Cross validation for hyperparameter selection"
]
},
{
"cell_type": "code",
"execution_count": 17,
"metadata": {
"collapsed": false
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Help on function cross_validation_folds in module mhcflurry.class1_allele_specific.cross_validation:\n",
"\n",
"cross_validation_folds(train_data, alleles=None, n_folds=3, drop_similar_peptides=False, imputer=None, impute_kwargs={'min_observations_per_allele': 2, 'min_observations_per_peptide': 2}, parallel_backend=None)\n",
Tim O'Donnell
committed
" Split a AffinityMeasurementDataset into n_folds cross validation folds for each allele,\n",
" optionally performing imputation.\n",
" \n",
" Parameters\n",
" -----------\n",
Tim O'Donnell
committed
" train_data : mhcflurry.AffinityMeasurementDataset\n",
" \n",
" alleles : string list, optional\n",
" Alleles to run cross validation on. Default: all alleles in\n",
" train_data.\n",
" \n",
" n_folds : int, optional\n",
" Number of cross validation folds for each allele.\n",
" \n",
" drop_similar_peptides : boolean, optional\n",
" For each fold, remove peptides from the test data that are similar\n",
" to peptides in the train data. Similarity is defined as in the\n",
" similar_peptides function.\n",
" \n",
" imputer : fancyimpute.Solver, optional\n",
" Imputer to use. If not specified, no imputation is done.\n",
" \n",
" impute_kwargs : dict, optional\n",
Tim O'Donnell
committed
" Additional kwargs to pass to mhcflurry.AffinityMeasurementDataset.impute_missing_values.\n",
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
" \n",
" parallel_backend : mhcflurry.parallelism.ParallelBackend, optional\n",
" Futures implementation to use for running on multiple threads,\n",
" processes, or nodes\n",
" \n",
" Returns\n",
" -----------\n",
" list of AlleleSpecificTrainTestFold of length num alleles * n_folds\n",
"\n"
]
}
],
"source": [
"help(mhcflurry.class1_allele_specific.cross_validation.cross_validation_folds)"
]
},
{
"cell_type": "code",
"execution_count": 18,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
Tim O'Donnell
committed
"[AlleleSpecificTrainTestFold(allele='HLA-A3301', train=AffinityMeasurementDataset(n=2026, alleles=['HLA-A3301']), imputed_train=None, test=AffinityMeasurementDataset(n=1014, alleles=['HLA-A3301'])),\n",
" AlleleSpecificTrainTestFold(allele='HLA-A3301', train=AffinityMeasurementDataset(n=2027, alleles=['HLA-A3301']), imputed_train=None, test=AffinityMeasurementDataset(n=1013, alleles=['HLA-A3301'])),\n",
" AlleleSpecificTrainTestFold(allele='HLA-A3301', train=AffinityMeasurementDataset(n=2027, alleles=['HLA-A3301']), imputed_train=None, test=AffinityMeasurementDataset(n=1013, alleles=['HLA-A3301']))]"
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
]
},
"execution_count": 18,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"folds = mhcflurry.class1_allele_specific.cross_validation.cross_validation_folds(train_data)\n",
"folds"
]
},
{
"cell_type": "code",
"execution_count": 19,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/plain": [
"{'activation': 'tanh',\n",
" 'batch_normalization': True,\n",
" 'batch_size': 128,\n",
" 'dropout_probability': 0.0,\n",
" 'embedding_output_dim': 32,\n",
" 'fraction_negative': 0.0,\n",
" 'impute': False,\n",
" 'init': 'glorot_uniform',\n",
" 'kmer_size': 9,\n",
" 'layer_sizes': [64],\n",
" 'loss': 'mse',\n",
" 'max_ic50': 50000.0,\n",
" 'n_training_epochs': 250,\n",
" 'optimizer': 'rmsprop',\n",
" 'output_activation': 'sigmoid',\n",
" 'pretrain_decay': 'numpy.exp(-epoch)'}"
]
},
"execution_count": 19,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"# Take a look at what hyperparameters are available for searching over.\n",
"mhcflurry.class1_allele_specific.train.HYPERPARAMETER_DEFAULTS.defaults"
]
},
{
"cell_type": "code",
"execution_count": 20,
"metadata": {
"collapsed": false
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Searching over 2 models.\n",
"First model: \n",
Tim O'Donnell
committed
"{'kmer_size': 9, 'n_training_epochs': 250, 'loss': 'mse', 'max_ic50': 50000.0, 'fraction_negative': 0.1, 'layer_sizes': [8], 'batch_size': 128, 'activation': 'tanh', 'init': 'glorot_uniform', 'dropout_probability': 0.0, 'pretrain_decay': 'numpy.exp(-epoch)', 'optimizer': 'rmsprop', 'impute': False, 'embedding_output_dim': 32, 'batch_normalization': True, 'output_activation': 'sigmoid'}\n"
]
}
],
"source": [
"models_to_search = mhcflurry.class1_allele_specific.train.HYPERPARAMETER_DEFAULTS.models_grid(\n",
" fraction_negative=[.1],\n",
" layer_sizes=[[8], [12]])\n",
"print(\"Searching over %d models.\" % len(models_to_search))\n",
"print(\"First model: \\n%s\" % models_to_search[0])"
]
},
{
"cell_type": "code",
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
"metadata": {
"collapsed": false
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Help on function train_across_models_and_folds in module mhcflurry.class1_allele_specific.train:\n",
"\n",
"train_across_models_and_folds(folds, model_descriptions, cartesian_product_of_folds_and_models=True, return_predictors=False, folds_per_task=1, parallel_backend=None)\n",
" Train and optionally test any number of models across any number of folds.\n",
" \n",
" Parameters\n",
" -----------\n",
" folds : list of AlleleSpecificTrainTestFold\n",
" \n",
" model_descriptions : list of dict\n",
" Models to test\n",
" \n",
" cartesian_product_of_folds_and_models : boolean, optional\n",
" If true, then a predictor is treained for each fold and model\n",
" description.\n",
" If false, then len(folds) must equal len(model_descriptions), and\n",
" the i'th model is trained on the i'th fold.\n",
" \n",
" return_predictors : boolean, optional\n",
" Include the trained predictors in the result.\n",
" \n",
" parallel_backend : mhcflurry.parallelism.ParallelBackend, optional\n",
" Futures implementation to use for running on multiple threads,\n",
" processes, or nodes\n",
" \n",
" Returns\n",
" -----------\n",
" pandas.DataFrame\n",
"\n"
]
}
],
"source": [
"help(mhcflurry.class1_allele_specific.train.train_across_models_and_folds)"
]
},
{
"cell_type": "code",
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
"outputs": [
{
"data": {
"text/html": [
"<div>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>allele</th>\n",
" <th>fold_num</th>\n",
" <th>model_num</th>\n",
" <th>train_size</th>\n",
" <th>test_size</th>\n",
" <th>imputed_train_size</th>\n",
" <th>train_tau</th>\n",
" <th>train_f1</th>\n",
" <th>train_auc</th>\n",
" <th>test_tau</th>\n",
" <th>...</th>\n",
" <th>model_batch_size</th>\n",
" <th>model_activation</th>\n",
" <th>model_init</th>\n",
" <th>model_dropout_probability</th>\n",
" <th>model_pretrain_decay</th>\n",
" <th>model_optimizer</th>\n",
" <th>model_impute</th>\n",
" <th>model_embedding_output_dim</th>\n",
" <th>model_batch_normalization</th>\n",
" <th>model_output_activation</th>\n",