-
Notifications
You must be signed in to change notification settings - Fork 0
/
results.html
613 lines (574 loc) · 33.3 KB
/
results.html
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
231
232
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
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
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
338
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
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
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
549
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
611
612
613
<!DOCTYPE html>
<html lang="" xml:lang="">
<head>
<meta charset="utf-8" />
<meta http-equiv="X-UA-Compatible" content="IE=edge" />
<title>3 Result and Discussion | Network analysis approach using morphological profiling of chemical perturbation</title>
<meta name="description" content="Exploring the intersection of graph representation learning and cell profiling" />
<meta name="generator" content="bookdown 0.33 and GitBook 2.6.7" />
<meta property="og:title" content="3 Result and Discussion | Network analysis approach using morphological profiling of chemical perturbation" />
<meta property="og:type" content="book" />
<meta property="og:description" content="Exploring the intersection of graph representation learning and cell profiling" />
<meta name="twitter:card" content="summary" />
<meta name="twitter:title" content="3 Result and Discussion | Network analysis approach using morphological profiling of chemical perturbation" />
<meta name="twitter:description" content="Exploring the intersection of graph representation learning and cell profiling" />
<meta name="author" content="Nima Chamyani" />
<meta name="date" content="2023-07-04" />
<meta name="viewport" content="width=device-width, initial-scale=1" />
<meta name="apple-mobile-web-app-capable" content="yes" />
<meta name="apple-mobile-web-app-status-bar-style" content="black" />
<link rel="prev" href="methods.html"/>
<link rel="next" href="references.html"/>
<script src="libs/jquery-3.6.0/jquery-3.6.0.min.js"></script>
<script src="https://cdn.jsdelivr.net/npm/[email protected]/dist/fuse.min.js"></script>
<link href="libs/gitbook-2.6.7/css/style.css" rel="stylesheet" />
<link href="libs/gitbook-2.6.7/css/plugin-table.css" rel="stylesheet" />
<link href="libs/gitbook-2.6.7/css/plugin-bookdown.css" rel="stylesheet" />
<link href="libs/gitbook-2.6.7/css/plugin-highlight.css" rel="stylesheet" />
<link href="libs/gitbook-2.6.7/css/plugin-search.css" rel="stylesheet" />
<link href="libs/gitbook-2.6.7/css/plugin-fontsettings.css" rel="stylesheet" />
<link href="libs/gitbook-2.6.7/css/plugin-clipboard.css" rel="stylesheet" />
<link href="libs/anchor-sections-1.1.0/anchor-sections.css" rel="stylesheet" />
<link href="libs/anchor-sections-1.1.0/anchor-sections-hash.css" rel="stylesheet" />
<script src="libs/anchor-sections-1.1.0/anchor-sections.js"></script>
<style type="text/css">
pre > code.sourceCode { white-space: pre; position: relative; }
pre > code.sourceCode > span { display: inline-block; line-height: 1.25; }
pre > code.sourceCode > span:empty { height: 1.2em; }
.sourceCode { overflow: visible; }
code.sourceCode > span { color: inherit; text-decoration: inherit; }
pre.sourceCode { margin: 0; }
@media screen {
div.sourceCode { overflow: auto; }
}
@media print {
pre > code.sourceCode { white-space: pre-wrap; }
pre > code.sourceCode > span { text-indent: -5em; padding-left: 5em; }
}
pre.numberSource code
{ counter-reset: source-line 0; }
pre.numberSource code > span
{ position: relative; left: -4em; counter-increment: source-line; }
pre.numberSource code > span > a:first-child::before
{ content: counter(source-line);
position: relative; left: -1em; text-align: right; vertical-align: baseline;
border: none; display: inline-block;
-webkit-touch-callout: none; -webkit-user-select: none;
-khtml-user-select: none; -moz-user-select: none;
-ms-user-select: none; user-select: none;
padding: 0 4px; width: 4em;
color: #aaaaaa;
}
pre.numberSource { margin-left: 3em; border-left: 1px solid #aaaaaa; padding-left: 4px; }
div.sourceCode
{ }
@media screen {
pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
}
code span.al { color: #ff0000; font-weight: bold; } /* Alert */
code span.an { color: #60a0b0; font-weight: bold; font-style: italic; } /* Annotation */
code span.at { color: #7d9029; } /* Attribute */
code span.bn { color: #40a070; } /* BaseN */
code span.bu { color: #008000; } /* BuiltIn */
code span.cf { color: #007020; font-weight: bold; } /* ControlFlow */
code span.ch { color: #4070a0; } /* Char */
code span.cn { color: #880000; } /* Constant */
code span.co { color: #60a0b0; font-style: italic; } /* Comment */
code span.cv { color: #60a0b0; font-weight: bold; font-style: italic; } /* CommentVar */
code span.do { color: #ba2121; font-style: italic; } /* Documentation */
code span.dt { color: #902000; } /* DataType */
code span.dv { color: #40a070; } /* DecVal */
code span.er { color: #ff0000; font-weight: bold; } /* Error */
code span.ex { } /* Extension */
code span.fl { color: #40a070; } /* Float */
code span.fu { color: #06287e; } /* Function */
code span.im { color: #008000; font-weight: bold; } /* Import */
code span.in { color: #60a0b0; font-weight: bold; font-style: italic; } /* Information */
code span.kw { color: #007020; font-weight: bold; } /* Keyword */
code span.op { color: #666666; } /* Operator */
code span.ot { color: #007020; } /* Other */
code span.pp { color: #bc7a00; } /* Preprocessor */
code span.sc { color: #4070a0; } /* SpecialChar */
code span.ss { color: #bb6688; } /* SpecialString */
code span.st { color: #4070a0; } /* String */
code span.va { color: #19177c; } /* Variable */
code span.vs { color: #4070a0; } /* VerbatimString */
code span.wa { color: #60a0b0; font-weight: bold; font-style: italic; } /* Warning */
</style>
<style type="text/css">
/* Used with Pandoc 2.11+ new --citeproc when CSL is used */
div.csl-bib-body { }
div.csl-entry {
clear: both;
}
.hanging div.csl-entry {
margin-left:2em;
text-indent:-2em;
}
div.csl-left-margin {
min-width:2em;
float:left;
}
div.csl-right-inline {
margin-left:2em;
padding-left:1em;
}
div.csl-indent {
margin-left: 2em;
}
</style>
<link rel="stylesheet" href="style.css" type="text/css" />
</head>
<body>
<div class="book without-animation with-summary font-size-2 font-family-1" data-basepath=".">
<div class="book-summary">
<nav role="navigation">
<ul class="summary">
<li class="chapter" data-level="" data-path="index.html"><a href="index.html"><i class="fa fa-check"></i>Intersecting Graph Representation Learning and Cell Profiling: A Novel Approach to Analyzing Complex Biomedical Data</a>
<ul>
<li class="chapter" data-level="" data-path="index.html"><a href="index.html#aim"><i class="fa fa-check"></i>Aim</a></li>
<li class="chapter" data-level="" data-path="index.html"><a href="index.html#what-can-be-found-in-this-document"><i class="fa fa-check"></i>What can be found in this document?</a></li>
</ul></li>
<li class="chapter" data-level="1" data-path="intro.html"><a href="intro.html"><i class="fa fa-check"></i><b>1</b> Introduction</a>
<ul>
<li class="chapter" data-level="1.1" data-path="intro.html"><a href="intro.html#graphs"><i class="fa fa-check"></i><b>1.1</b> Graphs</a></li>
<li class="chapter" data-level="1.2" data-path="intro.html"><a href="intro.html#graph-representation-learning"><i class="fa fa-check"></i><b>1.2</b> Graph representation learning</a></li>
<li class="chapter" data-level="1.3" data-path="intro.html"><a href="intro.html#cell-profiling"><i class="fa fa-check"></i><b>1.3</b> Cell profiling</a></li>
</ul></li>
<li class="chapter" data-level="2" data-path="methods.html"><a href="methods.html"><i class="fa fa-check"></i><b>2</b> Methods and Materials</a>
<ul>
<li class="chapter" data-level="2.1" data-path="methods.html"><a href="methods.html#data-preprocessing"><i class="fa fa-check"></i><b>2.1</b> Data Preprocessing</a>
<ul>
<li class="chapter" data-level="2.1.1" data-path="methods.html"><a href="methods.html#covid-19-cell-profilling-data"><i class="fa fa-check"></i><b>2.1.1</b> COVID-19 Cell profilling Data</a>
<ul>
<li class="chapter" data-level="2.1.1.1" data-path="methods.html"><a href="methods.html#normalization"><i class="fa fa-check"></i><b>2.1.1.1</b> Normalization</a></li>
<li class="chapter" data-level="2.1.1.2" data-path="methods.html"><a href="methods.html#dimensionality-reduction"><i class="fa fa-check"></i><b>2.1.1.2</b> Dimensionality Reduction</a></li>
<li class="chapter" data-level="2.1.1.3" data-path="methods.html"><a href="methods.html#development-of-a-binary-classification-of-data"><i class="fa fa-check"></i><b>2.1.1.3</b> Development of a binary classification of data</a></li>
</ul></li>
<li class="chapter" data-level="2.1.2" data-path="methods.html"><a href="methods.html#compound-protein-and-pathway-data-aggregation"><i class="fa fa-check"></i><b>2.1.2</b> Compound, Protein and Pathway Data Aggregation</a></li>
<li class="chapter" data-level="2.1.3" data-path="methods.html"><a href="methods.html#featurizing-the-biomedical-entities"><i class="fa fa-check"></i><b>2.1.3</b> Featurizing the Biomedical Entities</a>
<ul>
<li class="chapter" data-level="2.1.3.1" data-path="methods.html"><a href="methods.html#featurizing-compounds"><i class="fa fa-check"></i><b>2.1.3.1</b> Featurizing Compounds</a></li>
<li class="chapter" data-level="2.1.3.2" data-path="methods.html"><a href="methods.html#featurizing-proteins"><i class="fa fa-check"></i><b>2.1.3.2</b> Featurizing Proteins</a></li>
<li class="chapter" data-level="2.1.3.3" data-path="methods.html"><a href="methods.html#featurizing-pathways"><i class="fa fa-check"></i><b>2.1.3.3</b> Featurizing Pathways</a></li>
</ul></li>
<li class="chapter" data-level="2.1.4" data-path="methods.html"><a href="methods.html#covid-19-bio-graph"><i class="fa fa-check"></i><b>2.1.4</b> COVID-19 Bio-Graph</a></li>
<li class="chapter" data-level="2.1.5" data-path="methods.html"><a href="methods.html#representing-chemical-molecules-as-graph"><i class="fa fa-check"></i><b>2.1.5</b> Representing Chemical Molecules as Graph</a></li>
</ul></li>
<li class="chapter" data-level="2.2" data-path="methods.html"><a href="methods.html#models"><i class="fa fa-check"></i><b>2.2</b> Models</a>
<ul>
<li class="chapter" data-level="2.2.1" data-path="methods.html"><a href="methods.html#graph-level-molecular-predictor-glmp"><i class="fa fa-check"></i><b>2.2.1</b> Graph-Level Molecular Predictor (GLMP)</a></li>
<li class="chapter" data-level="2.2.2" data-path="methods.html"><a href="methods.html#bio-graph-integrative-classifierregressor-biogicbiogir"><i class="fa fa-check"></i><b>2.2.2</b> Bio-Graph Integrative Classifier/Regressor (BioGIC/BioGIR)</a>
<ul>
<li class="chapter" data-level="2.2.2.1" data-path="methods.html"><a href="methods.html#classificationregression"><i class="fa fa-check"></i><b>2.2.2.1</b> Classification/Regression</a></li>
<li class="chapter" data-level="2.2.2.2" data-path="methods.html"><a href="methods.html#predicting-joint-effect-of-nodes-chemical-combination"><i class="fa fa-check"></i><b>2.2.2.2</b> Predicting joint effect of nodes (Chemical Combination)</a></li>
</ul></li>
<li class="chapter" data-level="2.2.3" data-path="methods.html"><a href="methods.html#optimized-molecular-graph-generator-omg"><i class="fa fa-check"></i><b>2.2.3</b> Optimized Molecular Graph Generator (OMG)</a></li>
</ul></li>
<li class="chapter" data-level="2.3" data-path="methods.html"><a href="methods.html#model-validation-and-optimization"><i class="fa fa-check"></i><b>2.3</b> Model Validation and Optimization</a></li>
<li class="chapter" data-level="2.4" data-path="methods.html"><a href="methods.html#model-enhancement"><i class="fa fa-check"></i><b>2.4</b> Model Enhancement</a></li>
<li class="chapter" data-level="2.5" data-path="methods.html"><a href="methods.html#data-acquisition-software-and-libraries"><i class="fa fa-check"></i><b>2.5</b> Data acquisition, software and libraries</a></li>
</ul></li>
<li class="chapter" data-level="3" data-path="results.html"><a href="results.html"><i class="fa fa-check"></i><b>3</b> Result and Discussion</a>
<ul>
<li class="chapter" data-level="3.1" data-path="results.html"><a href="results.html#regressionclassification-performance"><i class="fa fa-check"></i><b>3.1</b> Regression/Classification Performance</a></li>
<li class="chapter" data-level="3.2" data-path="results.html"><a href="results.html#covid-19-biograph-topology"><i class="fa fa-check"></i><b>3.2</b> COVID-19 BioGraph Topology</a></li>
<li class="chapter" data-level="3.3" data-path="results.html"><a href="results.html#combination-prediction"><i class="fa fa-check"></i><b>3.3</b> Combination Prediction</a></li>
<li class="chapter" data-level="3.4" data-path="results.html"><a href="results.html#molecule-generation"><i class="fa fa-check"></i><b>3.4</b> Molecule Generation</a></li>
</ul></li>
<li class="chapter" data-level="" data-path="references.html"><a href="references.html"><i class="fa fa-check"></i>References</a></li>
</ul>
</nav>
</div>
<div class="book-body">
<div class="body-inner">
<div class="book-header" role="navigation">
<h1>
<i class="fa fa-circle-o-notch fa-spin"></i><a href="./">Network analysis approach using morphological profiling of chemical perturbation</a>
</h1>
</div>
<div class="page-wrapper" tabindex="-1" role="main">
<div class="page-inner">
<section class="normal" id="section-">
<div id="results" class="section level1 hasAnchor" number="3">
<h1><span class="header-section-number">3</span> Result and Discussion<a href="results.html#results" class="anchor-section" aria-label="Anchor link to header"></a></h1>
<div id="regressionclassification-performance" class="section level2 hasAnchor" number="3.1">
<h2><span class="header-section-number">3.1</span> Regression/Classification Performance<a href="results.html#regressionclassification-performance" class="anchor-section" aria-label="Anchor link to header"></a></h2>
<p>The performance of multiple machine learning models, such as the GLMP and BioGIP, was assessed through various tasks. Detailed results can be found in Appendix A.</p>
<p>A variety of conventional models such as Gradient Boosting (GBoost), Random Forest (RF), K-Nearest Neighbors, Decision Tree, Multi-layer Perceptron (MLP), Support Vector Machines Classifier/Regressor (SVC/SVR), AdaBoost, Gaussian Naive Bayes/Gaussian Process, Stochastic Gradient Descent (SGD), and Ridge Regression were included for comparison. While these models exhibited marginally better performance than the conventional models, it was clear that improvements were needed in their ability to perform regression on the PCA1 value. Currently, these models are not optimally suitable for regression tasks (Comparative Performance of Models Table).</p>
<p style="text-align: justify; text-align-last: left; font-size: 12px;">
Comparative Performance of Models on PCA1 Classification and Regression Tasks: This table presents the best-observed performance of GLMP and BioGIP models, contrasted with several conventional models. For input, conventional models and GLMP use a feature vector encompassing structural and physicochemical properties of molecules. GLMP further integrates this with global graph presentations of molecules. BioGIP employs distinct node features (chemical, protein, and pathways) and their links. When GLMP and BioGIP are connected, the chemical feature becomes the last layer of the GLMP model.
</p>
<div style="font-size: 53%;">
<table>
<colgroup>
<col width="12%" />
<col width="10%" />
<col width="11%" />
<col width="5%" />
<col width="3%" />
<col width="8%" />
<col width="9%" />
<col width="3%" />
<col width="5%" />
<col width="6%" />
<col width="13%" />
<col width="3%" />
<col width="4%" />
</colgroup>
<thead>
<tr class="header">
<th></th>
<th align="center"><strong>GLMP</strong></th>
<th align="center"><strong>BioGIP</strong></th>
<th align="center">GBoost</th>
<th align="center">RF</th>
<th align="center">KNNeighbors</th>
<th align="center">DecisionTree</th>
<th align="center">MLP</th>
<th align="center">SVC/SVR</th>
<th align="center">AdaBoost</th>
<th align="center">Gaussian NB/Process</th>
<th align="center">SGD</th>
<th align="center">Ridge</th>
</tr>
</thead>
<tbody>
<tr class="odd">
<td>AUC-ROC</td>
<td align="center">0.63</td>
<td align="center"><strong>0.75</strong></td>
<td align="center">0.57</td>
<td align="center">0.5</td>
<td align="center">0.52</td>
<td align="center">0.58</td>
<td align="center">0.57</td>
<td align="center">0.56</td>
<td align="center">0.5</td>
<td align="center">0.62</td>
<td align="center">0.6</td>
<td align="center"></td>
</tr>
<tr class="even">
<td>Balanced Accuracy</td>
<td align="center">0.58</td>
<td align="center"><strong>0.72</strong></td>
<td align="center">0.57</td>
<td align="center">0.5</td>
<td align="center">0.52</td>
<td align="center">0.58</td>
<td align="center">0.57</td>
<td align="center">0.56</td>
<td align="center">0.5</td>
<td align="center">0.62</td>
<td align="center">0.6</td>
<td align="center"></td>
</tr>
<tr class="odd">
<td>Recall</td>
<td align="center">0.26</td>
<td align="center"><strong>0.75</strong></td>
<td align="center">0.19</td>
<td align="center">0</td>
<td align="center">0.04</td>
<td align="center">0.19</td>
<td align="center">0.15</td>
<td align="center">0.15</td>
<td align="center">0</td>
<td align="center">0.37</td>
<td align="center">0.22</td>
<td align="center"></td>
</tr>
<tr class="even">
<td>Precision</td>
<td align="center">0.54</td>
<td align="center"><strong>0.58</strong></td>
<td align="center">0.11</td>
<td align="center">0</td>
<td align="center">0.17</td>
<td align="center">0.17</td>
<td align="center">0.25</td>
<td align="center">0.18</td>
<td align="center">0</td>
<td align="center">0.07</td>
<td align="center">0.19</td>
<td align="center"></td>
</tr>
<tr class="odd">
<td>F1</td>
<td align="center">0.36</td>
<td align="center"><strong>0.60</strong></td>
<td align="center">0.14</td>
<td align="center">0</td>
<td align="center">0.06</td>
<td align="center">0.18</td>
<td align="center">0.19</td>
<td align="center">0.16</td>
<td align="center">0</td>
<td align="center">0.12</td>
<td align="center">0.2</td>
<td align="center"></td>
</tr>
<tr class="even">
<td><span class="math inline">\(R^2\)</span></td>
<td align="center"><strong>0.16</strong></td>
<td align="center">0.07</td>
<td align="center">0</td>
<td align="center">0.03</td>
<td align="center">0.04</td>
<td align="center">0</td>
<td align="center">$<$0</td>
<td align="center">0.05</td>
<td align="center">0.03</td>
<td align="center">$<$0</td>
<td align="center"></td>
<td align="center">$<$0</td>
</tr>
<tr class="odd">
<td>mse</td>
<td align="center"><strong>4.89</strong></td>
<td align="center">5.25</td>
<td align="center">5.56</td>
<td align="center">5.36</td>
<td align="center">5.31</td>
<td align="center">5.54</td>
<td align="center">6.24</td>
<td align="center">5.27</td>
<td align="center">5.38</td>
<td align="center">6.36</td>
<td align="center"></td>
<td align="center">7.49</td>
</tr>
<tr class="even">
<td>rmse</td>
<td align="center"><strong>2.21</strong></td>
<td align="center">2.29</td>
<td align="center">2.36</td>
<td align="center">2.32</td>
<td align="center">2.31</td>
<td align="center">2.35</td>
<td align="center">2.5</td>
<td align="center">2.3</td>
<td align="center">2.32</td>
<td align="center">2.52</td>
<td align="center"></td>
<td align="center">2.74</td>
</tr>
</tbody>
</table>
</div>
<p>In the classification tasks, an interesting pattern was observed. The BioGIP model demonstrated superior performance across most metrics, including AUC-ROC, Balanced Accuracy, Recall, Precision, and F1-Score. However, when it came to the regression tasks, the GLMP and an enhanced version of GLMP, GLMP-PreGIN, displayed a higher R<span class="math inline">\(^2\)</span> value and lower mean squared error (MSE) and root mean squared error (RMSE), potentially suggesting a better fit of the model to the data.</p>
<p>Further experiment was conducted by introducing different GLMP and BioGIP model enhancement approaches. Here, the GLMP-BioGIP model yielded the best performance across both classification and regression tasks, as indicated by the metrics (Performance of Enhanced Models on Classification and Regression Table).</p>
<p style="text-align: justify; text-align-last: left; font-size: 12px;">
Performance of Enhanced Models on Classification and Regression Tasks: This table compares the performance of different model enhancement approaches on GLMP and BioGIP models.
</p>
<div style="font-size: 60%;">
<table>
<colgroup>
<col width="16%" />
<col width="5%" />
<col width="19%" />
<col width="13%" />
<col width="13%" />
<col width="15%" />
<col width="15%" />
</colgroup>
<thead>
<tr class="header">
<th align="left"></th>
<th align="center">GLMP</th>
<th align="center">GLMP-PreGIN</th>
<th align="center">BioGIP<span class="math inline">\(_{\text{seq}}\)</span></th>
<th align="center">BioGIP<span class="math inline">\(_{\text{cat}}\)</span></th>
<th align="center">BioGIP<span class="math inline">\(_{\text{res}}\)</span></th>
<th align="center">GLMP-BioGIP</th>
</tr>
</thead>
<tbody>
<tr class="odd">
<td align="left">AUC-ROC</td>
<td align="center">0.61</td>
<td align="center">0.63</td>
<td align="center">0.75</td>
<td align="center">0.69</td>
<td align="center">0.59</td>
<td align="center"><strong>0.78</strong></td>
</tr>
<tr class="even">
<td align="left">Balanced Accuracy</td>
<td align="center">0.56</td>
<td align="center">0.58</td>
<td align="center">0.72</td>
<td align="center">0.68</td>
<td align="center">0.55</td>
<td align="center"><strong>0.71</strong></td>
</tr>
<tr class="odd">
<td align="left">Recall</td>
<td align="center">0.21</td>
<td align="center">0.26</td>
<td align="center">0.75</td>
<td align="center">0.62</td>
<td align="center">0.53</td>
<td align="center"><strong>0.77</strong></td>
</tr>
<tr class="even">
<td align="left">Precision</td>
<td align="center">0.18</td>
<td align="center">0.54</td>
<td align="center">0.58</td>
<td align="center">0.53</td>
<td align="center">0.51</td>
<td align="center"><strong>0.62</strong></td>
</tr>
<tr class="odd">
<td align="left">F1</td>
<td align="center">0.2</td>
<td align="center">0.36</td>
<td align="center">0.6</td>
<td align="center">0.51</td>
<td align="center">0.44</td>
<td align="center"><strong>0.66</strong></td>
</tr>
<tr class="even">
<td align="left"><span class="math inline">\(R^2\)</span></td>
<td align="center">0.01</td>
<td align="center">0.16</td>
<td align="center">0.07</td>
<td align="center">0.03</td>
<td align="center">0</td>
<td align="center"><strong>0.17</strong></td>
</tr>
<tr class="odd">
<td align="left">mse</td>
<td align="center">5.49</td>
<td align="center">4.89</td>
<td align="center">5.26</td>
<td align="center">5.35</td>
<td align="center">5.6</td>
<td align="center"><strong>4.87</strong></td>
</tr>
<tr class="even">
<td align="left">rmse</td>
<td align="center">2.34</td>
<td align="center">2.21</td>
<td align="center">2.29</td>
<td align="center">2.31</td>
<td align="center">2.37</td>
<td align="center"><strong>2.20</strong></td>
</tr>
</tbody>
</table>
</div>
</div>
<div id="covid-19-biograph-topology" class="section level2 hasAnchor" number="3.2">
<h2><span class="header-section-number">3.2</span> COVID-19 BioGraph Topology<a href="results.html#covid-19-biograph-topology" class="anchor-section" aria-label="Anchor link to header"></a></h2>
<p>The COVID-19 BioGraph is a comprehensive network of several key components - chemicals, proteins, and biological pathways. These elements are intricately interconnected, forming a complex topology that underscores the dynamism of biological systems.</p>
<p>Of the 4,293 chemicals represented in the BioGraph, 3,711 are known to interact with at least one protein. This highlights chemicals’ crucial role in influencing protein function, indicating a high degree of interaction between these two entities.</p>
<p>The total number of proteins in the BioGraph is 16,733, with a striking 16,727 shown to interact with chemicals. The near-universal interaction between proteins and chemicals reinforces their integral role in maintaining and regulating biological processes.</p>
<div class="float">
<img src="assets/biograph_module.jpg" alt="BioGraph" />
<div class="figcaption">BioGraph</div>
</div>
<p>The BioGraph also features 1,117 unique biological pathways. However, only 282 proteins are identified as governing these pathways, demonstrating a subset of proteins’ pivotal role in controlling and influencing various biological processes. The role of chemicals is again emphasized as it is found that 3,220 of them are linked to all biological pathways. This widespread involvement of chemicals in biological pathways indicates their significant influence on the overall functioning of biological processes.</p>
<p>A subgraph of the BioGraph, focusing on 136 active compounds, has been depicted for a more manageable analysis. There are three different types of nodes in this subgraph - green, pink, and blue - representing compounds, pathways, and proteins. The size of the green nodes represents the degree of the compounds, thus offering a visual representation of their connectivity within the network.</p>
<p>In the active subgraph, module detection has been employed to identify interconnected communities of compounds related to specific proteins and pathways. This method potentially reveals disparate mechanisms by highlighting unique compound-protein-pathway interactions.</p>
</div>
<div id="combination-prediction" class="section level2 hasAnchor" number="3.3">
<h2><span class="header-section-number">3.3</span> Combination Prediction<a href="results.html#combination-prediction" class="anchor-section" aria-label="Anchor link to header"></a></h2>
<p>Finding active chemical combinations within inactive compounds was obtained by implementing two distinct strategies aimed at predicting the joint effect of nodes, or chemical combinations, in the BioGraph. The methods used were the PCA1 range selection and the louvain community detection algorithm, both of which facilitated the selection of hypothetical inactive nodes that could potentially make an active combination.</p>
<p>In the first approach, an emphasis was placed on employing PCA1 values, specifically those ranging between 2.9 and 5, in selecting chemical compounds for combination. This particular range was chosen due to its proximity to the values exhibited by active compounds. The technique led to the selection of 141 nodes (molecules), thereby reducing the computational burden by limiting the potential compound combinations to a manageable 9,871 pairs.</p>
<p>In the second strategy, the louvain community detection algorithm was utilized to select promising inactive nodes. With its capability of revealing hidden patterns and reducing network complexity, this algorithm proved to be a potent alternative to PCA1-based compound selection. It yielded 159 compounds for combination and a total of 12,561 possible combinations.</p>
<div class="float">
<img src="assets/active_combo.png" alt="Active Combinations" />
<div class="figcaption">Active Combinations</div>
</div>
<p>The selected combinations were then tested using our most effective model, BioGIP-GLMP. As a result, 1519 combinations from the first approach (representing 15.4% of the total) and 1183 combinations from the second approach (amounting to 9.4% of the total) were identified as active, with a probability range of 0.5 to 1.</p>
<p>A number of these combinations are shown in the figure above. Notably, there was a significant overlap between the results of both strategies. Of the identified active combinations, 298 pairs were similar in both approaches, representing approximately 20% of the active combinations from the PCA1 range selection method and 25% from the louvain community detection approach.</p>
<p>A vital caveat to note in the PCA1 approach is that the number of inactive compounds within the selected range remained fixed. However, for the louvain community detection method, the number of identified communities and the population of active compounds within these communities varied depending on the chosen resolution. Furthermore, as the louvain community detection algorithm uses a random approach to identify modularity, the results differed with each iteration.</p>
<p>At resolution 1, between 11 to 16 distinct modules were typically identified, with active compounds scattered across half of them. A growing separation between active and inactive compounds within the communities was observed as the resolution was incrementally increased. Nevertheless, several mixed communities persisted. These mixed communities, containing both active and inactive compounds, offered intriguing possibilities as they contained inactive compounds that could potentially behave like active compounds within the class. Consequently, they were considered good candidates for combination.</p>
<p>Finally, an adjustment was set to level 7, with all communities possessing at least 20% active compound. These communities were consequently isolated. Subsequently, the inactive compounds for combination were segregated within these isolated communities.</p>
</div>
<div id="molecule-generation" class="section level2 hasAnchor" number="3.4">
<h2><span class="header-section-number">3.4</span> Molecule Generation<a href="results.html#molecule-generation" class="anchor-section" aria-label="Anchor link to header"></a></h2>
<p>As part of the OMG model, a suite of graph-based generative models was employed to generate optimized molecules, namely, the Graph Convolutional Policy Network (GCPN) and GraphAF. These models were founded on the premise of understanding the input graph’s inherent features, such as node types and edge types, which facilitated the generation of new graphs. The performance of these models was then evaluated using an ordinal regression model, an alternative to the less successful GLMP and BioGIP regressor models, which showed poor regression performance with an R<span class="math inline">\(^2\)</span> of 0.17 at their best.</p>
<p>The ordinal regression model was utilized because of its ability to overcome the inadequacies of the other models. It simplifies the problem of quantifying the goodness of a generated molecule, making it a suitable measure for the reinforcement learning paradigm used by OMG. The model segmented the PCA1 value into ordinal categories, facilitating a more straightforward interpretation of the order of goodness of the generated molecules.</p>
<p>The PCA1 value was divided into five categories as per the ordinal regression: (-12, -5), (-5, 2), (2, 9), (9, 16). The results showed that the model performed relatively well, with an overall accuracy of 56.36%. It was noted that the model performed best on molecules with PCA1 values within class 2 (-5, 2), yielding the highest f1-score. On closer inspection of the results, it was found that the compounds with PCA1 values above five were often classified within class 3 (2, 9), with a lesser number falling within class 4 (9, 16). However, a decision was made to count molecules with PCA1 values above nine active, not those above five, to ensure a stringent definition of an “active” compound during the training phase. This choice made the model more conservative, thus reducing the risk of generating false positives.</p>
<p>Regarding the specific attributes of the generated molecules, both GCPN and GraphAF offered unique benefits. Molecules generated by GraphAF were generally simpler and smaller, often incorporating atoms other than carbon, which could potentially lead to better solubility and decreased hydrophobicity. Conversely, GCPN tended to generate more complex molecules, indicating these models’ versatility in producing various molecular structures.</p>
<p>Overall, the ordinal regression model proved beneficial for evaluating the molecules generated by the OMG models. By adopting a conservative stance during training, the model mitigated the risk of producing false positives, thus offering potential avenues for future research in optimized molecular generation.</p>
</div>
</div>
</section>
</div>
</div>
</div>
<a href="methods.html" class="navigation navigation-prev " aria-label="Previous page"><i class="fa fa-angle-left"></i></a>
<a href="references.html" class="navigation navigation-next " aria-label="Next page"><i class="fa fa-angle-right"></i></a>
</div>
</div>
<script src="libs/gitbook-2.6.7/js/app.min.js"></script>
<script src="libs/gitbook-2.6.7/js/clipboard.min.js"></script>
<script src="libs/gitbook-2.6.7/js/plugin-search.js"></script>
<script src="libs/gitbook-2.6.7/js/plugin-sharing.js"></script>
<script src="libs/gitbook-2.6.7/js/plugin-fontsettings.js"></script>
<script src="libs/gitbook-2.6.7/js/plugin-bookdown.js"></script>
<script src="libs/gitbook-2.6.7/js/jquery.highlight.js"></script>
<script src="libs/gitbook-2.6.7/js/plugin-clipboard.js"></script>
<script>
gitbook.require(["gitbook"], function(gitbook) {
gitbook.start({
"sharing": {
"github": false,
"facebook": true,
"twitter": true,
"linkedin": false,
"weibo": false,
"instapaper": false,
"vk": false,
"whatsapp": false,
"all": ["facebook", "twitter", "linkedin", "weibo", "instapaper"]
},
"fontsettings": {
"theme": "white",
"family": "sans",
"size": 2
},
"edit": null,
"history": {
"link": null,
"text": null
},
"view": {
"link": null,
"text": null
},
"download": ["_main.pdf", "_main.epub", "_main.mobi"],
"search": {
"engine": "fuse",
"options": null
},
"toc": {
"collapse": "subsection"
},
"toolbar": {
"position": "fixed"
}
});
});
</script>
<!-- dynamically load mathjax for compatibility with self-contained -->
<script>
(function () {
var script = document.createElement("script");
script.type = "text/javascript";
var src = "true";
if (src === "" || src === "true") src = "https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.9/latest.js?config=TeX-MML-AM_CHTML";
if (location.protocol !== "file:")
if (/^https?:/.test(src))
src = src.replace(/^https?:/, '');
script.src = src;
document.getElementsByTagName("head")[0].appendChild(script);
})();
</script>
</body>
</html>