-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathtwoGroupsBackground.html
More file actions
401 lines (335 loc) · 14.5 KB
/
twoGroupsBackground.html
File metadata and controls
401 lines (335 loc) · 14.5 KB
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
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8" />
<meta name="generator" content="pandoc" />
<meta http-equiv="X-UA-Compatible" content="IE=EDGE" />
<meta name="author" content="Timothy Daley" />
<meta name="date" content="2020-12-27" />
<title>Background of two groups model</title>
<script src="site_libs/jquery-1.11.3/jquery.min.js"></script>
<meta name="viewport" content="width=device-width, initial-scale=1" />
<link href="site_libs/bootstrap-3.3.5/css/yeti.min.css" rel="stylesheet" />
<script src="site_libs/bootstrap-3.3.5/js/bootstrap.min.js"></script>
<script src="site_libs/bootstrap-3.3.5/shim/html5shiv.min.js"></script>
<script src="site_libs/bootstrap-3.3.5/shim/respond.min.js"></script>
<script src="site_libs/navigation-1.1/tabsets.js"></script>
<link href="site_libs/highlightjs-9.12.0/textmate.css" rel="stylesheet" />
<script src="site_libs/highlightjs-9.12.0/highlight.js"></script>
<link href="site_libs/anchor-sections-1.0/anchor-sections.css" rel="stylesheet" />
<script src="site_libs/anchor-sections-1.0/anchor-sections.js"></script>
<style type="text/css">
code{white-space: pre-wrap;}
span.smallcaps{font-variant: small-caps;}
span.underline{text-decoration: underline;}
div.column{display: inline-block; vertical-align: top; width: 50%;}
div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
ul.task-list{list-style: none;}
</style>
<style type="text/css">code{white-space: pre;}</style>
<style type="text/css">
pre:not([class]) {
background-color: white;
}
</style>
<script type="text/javascript">
if (window.hljs) {
hljs.configure({languages: []});
hljs.initHighlightingOnLoad();
if (document.readyState && document.readyState === "complete") {
window.setTimeout(function() { hljs.initHighlighting(); }, 0);
}
}
</script>
<style type="text/css">
h1 {
font-size: 34px;
}
h1.title {
font-size: 38px;
}
h2 {
font-size: 30px;
}
h3 {
font-size: 24px;
}
h4 {
font-size: 18px;
}
h5 {
font-size: 16px;
}
h6 {
font-size: 12px;
}
.table th:not([align]) {
text-align: left;
}
</style>
<style type = "text/css">
.main-container {
max-width: 940px;
margin-left: auto;
margin-right: auto;
}
code {
color: inherit;
background-color: rgba(0, 0, 0, 0.04);
}
img {
max-width:100%;
}
.tabbed-pane {
padding-top: 12px;
}
.html-widget {
margin-bottom: 20px;
}
button.code-folding-btn:focus {
outline: none;
}
summary {
display: list-item;
}
</style>
<style type="text/css">
/* padding for bootstrap navbar */
body {
padding-top: 45px;
padding-bottom: 40px;
}
/* offset scroll position for anchor links (for fixed navbar) */
.section h1 {
padding-top: 50px;
margin-top: -50px;
}
.section h2 {
padding-top: 50px;
margin-top: -50px;
}
.section h3 {
padding-top: 50px;
margin-top: -50px;
}
.section h4 {
padding-top: 50px;
margin-top: -50px;
}
.section h5 {
padding-top: 50px;
margin-top: -50px;
}
.section h6 {
padding-top: 50px;
margin-top: -50px;
}
.dropdown-submenu {
position: relative;
}
.dropdown-submenu>.dropdown-menu {
top: 0;
left: 100%;
margin-top: -6px;
margin-left: -1px;
border-radius: 0 6px 6px 6px;
}
.dropdown-submenu:hover>.dropdown-menu {
display: block;
}
.dropdown-submenu>a:after {
display: block;
content: " ";
float: right;
width: 0;
height: 0;
border-color: transparent;
border-style: solid;
border-width: 5px 0 5px 5px;
border-left-color: #cccccc;
margin-top: 5px;
margin-right: -10px;
}
.dropdown-submenu:hover>a:after {
border-left-color: #ffffff;
}
.dropdown-submenu.pull-left {
float: none;
}
.dropdown-submenu.pull-left>.dropdown-menu {
left: -100%;
margin-left: 10px;
border-radius: 6px 0 6px 6px;
}
</style>
<script>
// manage active state of menu based on current page
$(document).ready(function () {
// active menu anchor
href = window.location.pathname
href = href.substr(href.lastIndexOf('/') + 1)
if (href === "")
href = "index.html";
var menuAnchor = $('a[href="' + href + '"]');
// mark it active
menuAnchor.parent().addClass('active');
// if it's got a parent navbar menu mark it active as well
menuAnchor.closest('li.dropdown').addClass('active');
});
</script>
<!-- tabsets -->
<style type="text/css">
.tabset-dropdown > .nav-tabs {
display: inline-table;
max-height: 500px;
min-height: 44px;
overflow-y: auto;
background: white;
border: 1px solid #ddd;
border-radius: 4px;
}
.tabset-dropdown > .nav-tabs > li.active:before {
content: "";
font-family: 'Glyphicons Halflings';
display: inline-block;
padding: 10px;
border-right: 1px solid #ddd;
}
.tabset-dropdown > .nav-tabs.nav-tabs-open > li.active:before {
content: "";
border: none;
}
.tabset-dropdown > .nav-tabs.nav-tabs-open:before {
content: "";
font-family: 'Glyphicons Halflings';
display: inline-block;
padding: 10px;
border-right: 1px solid #ddd;
}
.tabset-dropdown > .nav-tabs > li.active {
display: block;
}
.tabset-dropdown > .nav-tabs > li > a,
.tabset-dropdown > .nav-tabs > li > a:focus,
.tabset-dropdown > .nav-tabs > li > a:hover {
border: none;
display: inline-block;
border-radius: 4px;
background-color: transparent;
}
.tabset-dropdown > .nav-tabs.nav-tabs-open > li {
display: block;
float: none;
}
.tabset-dropdown > .nav-tabs > li {
display: none;
}
</style>
<!-- code folding -->
</head>
<body>
<div class="container-fluid main-container">
<div class="navbar navbar-default navbar-fixed-top" role="navigation">
<div class="container">
<div class="navbar-header">
<button type="button" class="navbar-toggle collapsed" data-toggle="collapse" data-target="#navbar">
<span class="icon-bar"></span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
</button>
<a class="navbar-brand" href="index.html">HOME</a>
</div>
<div id="navbar" class="navbar-collapse collapse">
<ul class="nav navbar-nav">
<li>
<a href="FisherVsStouffer.html">Fisher vs Stouffer</a>
</li>
<li>
<a href="twoGroupsBackground.html">2 groups</a>
</li>
<li>
<a href="empiricalFDR.html">empirical estimation of FDRs</a>
</li>
</ul>
<ul class="nav navbar-nav navbar-right">
</ul>
</div><!--/.nav-collapse -->
</div><!--/.container -->
</div><!--/.navbar -->
<div class="fluid-row" id="header">
<h1 class="title toc-ignore">Background of two groups model</h1>
<h4 class="author">Timothy Daley</h4>
<h4 class="date">12/27/2020</h4>
</div>
<p>A common trope in genomics is the two groups model. You have a large group of measurement, at least in the thousands, and usually more. The objective of the analysis is to identify a group of “interesting” examples from the group of measurements. Some examples:</p>
<ul>
<li><p>In standard RNA-seq analysis of a treatment vs control experiment (e.g. cancerous vs non-cancerous, drug treated vs untreated), the goal of differential expression is to identify genes that are different in the treatment compared to control under the assumption that most gene expression is unchanged between the case and control;</p></li>
<li><p>In GWAS analysis the goal is to identify loci or SNPs that are different in the diseased population compared the the healthy/control populations under the assumption that most differences are small;</p></li>
<li><p>In outlier identification the goal is to identify measurements that are signficantly different from the large majority of the rest of the population.</p></li>
</ul>
<p>A fundamental assumptions of the two groups is that there are two groups in the set of measurements that follow different distribution and that one group makes up a large majority of the whole set. To concretely define the two groups model we follow <a href="https://pdfs.semanticscholar.org/93dd/1ad905da7b09568aaf7d04c3d325772d42fc.pdf">Efron, 2005</a>. Suppose that we have <span class="math inline">\(N\)</span> measurements with <span class="math inline">\(N\)</span> hypotheses <span class="math inline">\(H_{1}, \ldots, H_{N}\)</span>. In Efron’s case the measurements were 1-dimensional but we won’t restrict ourselves to this. Denote the measurements <span class="math inline">\(\boldsymbol{x}_{1}, \ldots, \boldsymbol{x}_{N}\)</span>. The assumption that there are two groups of measurements corresponds to assuming that the distribution of the measurements is a mixture distribution, e.g. <span class="math inline">\(\boldsymbol{x}_{i} \sim (1 - p) f_{0}( \boldsymbol{x}) + p f_{1}(\boldsymbol{x})\)</span> for some small but non-zero <span class="math inline">\(p\)</span> and <span class="math inline">\(f_{1}\)</span> sufficiently different from <span class="math inline">\(f_{0}\)</span>. <span class="math inline">\(p\)</span> is the probability that a randomly chosen hypothesis is not null.</p>
<p>For a given measurement <span class="math inline">\(\boldsymbol{x}\)</span> we define the local fdr as the posterior probability that the corresponding hypothesis is null as <span class="math display">\[
\text{fdr}(\boldsymbol{x}) = \Pr(\text{ hypothesis } i \text{ is null } | \boldsymbol{x}) = \frac{(1 - p) f_{0}(\boldsymbol{x})}{(1 - p) f_{0}(\boldsymbol{x}) + p f_{1}(\boldsymbol{x})}.
\]</span> In the 1-dimensional case we usually make the assumption that the interesting cases are large positive measurements, and then we can define the global FDR of measurement <span class="math inline">\(x\)</span> as the tail area (see equation 2.7 in <a href="https://pdfs.semanticscholar.org/93dd/1ad905da7b09568aaf7d04c3d325772d42fc.pdf">Efron, 2005</a>) <span class="math display">\[
\text{FDR}(x) = \Pr (\text{ hypothesis } i \text{ is null } | X \leq x) = \frac{ (1 - p) \int_{z \leq x} f_{0} (z) dz}{(1 - p) \int_{z \leq x} f_{0} (z) dz + p \int_{z \leq x} f_{1} (z) dz}.
\]</span></p>
<p>In the multidimensional case what would we consider more extreme than a measurement <span class="math inline">\(\boldsymbol{x}\)</span>? Well naturally we would take measurements that are less null, i.e. measurements with a smaller local false discovery rate.<br />
<span class="math display">\[
\text{FDR}(\boldsymbol{x}) = \frac{ (1 - p) \int_{\boldsymbol{z}: \text{fdr}(\boldsymbol{z}) \leq \text{fdr}(\boldsymbol{x})} f_{0} (\boldsymbol{z}) d \boldsymbol{z}}{(1 - p) \int_{\boldsymbol{z}: \text{fdr}(\boldsymbol{z}) \leq \text{fdr}(\boldsymbol{x})} f_{0} (\boldsymbol{z}) d \boldsymbol{z} + p \int_{\boldsymbol{z}: \text{fdr}(\boldsymbol{z}) \leq \text{fdr}(\boldsymbol{x})} f_{1} (\boldsymbol{z}) d \boldsymbol{z}}.
\]</span></p>
<div id="examples" class="section level1">
<h1>Examples</h1>
<div id="p-values" class="section level2">
<h2><span class="math inline">\(p\)</span>-values</h2>
<p>A pretty clear example is in <span class="math inline">\(p\)</span>-values, a common summary statistic of a hypothesis. A difference here is that there is a strong theoretical null distribution if the null hyotheseses are all correctly specified, namely a <span class="math inline">\(\text{Uniform}(0, 1)\)</span>. This leads to something like the figure below taken from Figure 1 of <a href="https://www.pnas.org/content/100/16/9440.full?__hstc=31177824.2a1e835c34ab7bf88e972fdd7a7debc8.1442188800044.1442188800045.1442188800046.1&__hssc=31177824.1.1442188800047&__hsfp=1314462730">Storey and Tibshirani, 2003</a>.</p>
<p><img src="F1.large.jpg" alt="A mixture of null and non-null p-values" width="400"/></p>
<p>The non-null group can be seen as those in excess of expected under the null in the left hand tail of the distribution. The known null allows for easy estimation of <span class="math inline">\(p\)</span> and <span class="math inline">\(f_{1}\)</span>.</p>
</div>
<div id="differential-expression" class="section level2">
<h2>Differential expression</h2>
<p>In differential expression analysis we assume that after normalization most gene effects are zero, then we look for log fold changes that are significantly different from 0. This is shown below, taken from Figure 4 in <a href="https://academic.oup.com/nar/article/43/7/e47/2414268">Ritchie et al, 2015</a>. The red points are the genes that were identified as significantly different from zero.</p>
<p><img src="differentialExpressionLogFoldChangeA.png" alt="Differential expression with log fold change" width="400"/></p>
</div>
<div id="high-throughput-screening-for-drug-discovery" class="section level2">
<h2>High throughput screening for drug discovery</h2>
<p>High throughput screening is usually the first step in identifying possible targets for druggable molecules, e.g. <a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3058157/">Hughes et al, 2011</a>. A large swath of molecules are assayed and measured for particular chemical properties, such as binding to a desired molecule. Then the outliers in the desired direction are chosen as possible hits for further in depth investigation, e.g. <a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4965338/">Shaby et al, 2016</a>.</p>
</div>
<div id="pooled-crispr-screens" class="section level2">
<h2>Pooled CRISPR screens</h2>
<p>The primary motivating application of the two groups model for me is pooled CRISPR screens. In such screens thousands of genes or loci are perturbed with CRISPRko (knockout), CRISPRi (inhibition), or CRISPRa (activation), and then selected for some phenotype. The genes with significant changes (positive or negative, depending on the application) from the rest are likely to be associated with the desired phenotype and are chosen for further investigation. One difference in this application is that there are several measurements per gene and these can show high variability in their effects, even for known positive hits. This lead me to suggest a hierarchical two groups model, shown below.</p>
<p><img src="CRISPhieRmixFig1.png" alt="CRISPhieRmix model" width="400"/></p>
</div>
</div>
</div>
<script>
// add bootstrap table styles to pandoc tables
function bootstrapStylePandocTables() {
$('tr.odd').parent('tbody').parent('table').addClass('table table-condensed');
}
$(document).ready(function () {
bootstrapStylePandocTables();
});
</script>
<!-- tabsets -->
<script>
$(document).ready(function () {
window.buildTabsets("TOC");
});
$(document).ready(function () {
$('.tabset-dropdown > .nav-tabs > li').click(function () {
$(this).parent().toggleClass('nav-tabs-open')
});
});
</script>
<!-- code folding -->
<!-- dynamically load mathjax for compatibility with self-contained -->
<script>
(function () {
var script = document.createElement("script");
script.type = "text/javascript";
script.src = "https://mathjax.rstudio.com/latest/MathJax.js?config=TeX-AMS-MML_HTMLorMML";
document.getElementsByTagName("head")[0].appendChild(script);
})();
</script>
</body>
</html>