-
Notifications
You must be signed in to change notification settings - Fork 8
/
causal_fmri.html
165 lines (124 loc) · 5.21 KB
/
causal_fmri.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
<!DOCTYPE html>
<html>
<head>
<title>Batch Causality</title>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8" />
<link rel="stylesheet" href="fonts/quadon/quadon.css">
<link rel="stylesheet" href="fonts/gentona/gentona.css">
<link rel="stylesheet" href="slides_style_i.css">
<script type="text/javascript" src="assets/plotly/plotly-latest.min.js"></script>
</head>
<body>
<textarea id="source">
<!-- TODO add slide numbers & maybe slide name -->
## Batch Effects are Causal Effects
Eric Bridgeford | PhD Student, Dept. Biostatistics, JHU
ericwb.me
---
## What does a batch effect look like?
- Example
- Data: $N$ participants, measure functional connectivity (FC)
- Goal: want to uncover whether there is a difference between males and females
---
## Batch Effect Example
<img src="images/causal_fmri/dataset1.png" style="height: 250px"/>
---
## Batch Effect Example
<center><img src="images/causal_fmri/dataset1.png" style="height: 250px"/>
<img src="images/causal_fmri/dataset2.png" style="height: 250px"/></center>
---
## Batch Effect Example
<center><img src="images/causal_fmri/dataset1.png" style="height: 250px"/>
<img src="images/causal_fmri/dataset2.png" style="height: 250px"/></center>
<center><img src="images/causal_fmri/agree.png" style="height: 250px"/></center>
---
## What happened?
- Datasets 1 and 2 differ in covariate distribution
<img src="images/causal_fmri/demo.png" style="height: 300px"/>
---
## Batch Effect Example -- Try 2
<newline> </newline>
<center><img src="images/causal_fmri/dataset1.png" style="height: 250px"/>
<img src="images/causal_fmri/dataset0.png" style="height: 250px"/></center>
<center><img src="images/causal_fmri/agree13.png" style="height: 250px"/></center>
---
## Procedure
- subset of CoRR Dataset
- $N>1,700$ individuals imaged across $24$ different datasets
- Estimate FC for all individuals with AAL parcellation
- all datasets have associated covariates (sex, age, continent)
- for all pairs of datasets $d_1, d_2$:
- Compute association of dataset with FC;
- Estimate causal effect of dataset with FC if such a comparison is possible;
- key aspect: comparison is, quite often, .ye[impossible]
---
### Adjustment for Covariate Distribution Still yields Batch Effects
- $\frac{24\cdot23}{2} = 276$ possible Comparisons
- $170$ pairs differ in covariate distribution
- $106$ pairs have similar covariate distribution
<newline> </newline>
<newline> </newline>
<center><img src="images/causal_fmri/before.png" style="height: 250px"/></center>
---
### Batch Effects and Independence Effects can be Eliminated through ComBat
<center><img src="images/causal_fmri/after.png" style="height: 250px"/></center>
---
### ComBat preserves between-individual Signal
- are we preserving signal that .ye[should] be present between individuals?
- Problem: site correction could "wash out" inter-individual comparisons
| Signal | $\%$ Raw Data | $\%$ After ComBat |
| :--- | :--- | :--- |
| Sex$^\dagger$ | $\frac{19}{22}$ | $\frac{18}{22}$
| Age | $\frac{18}{24}$ | $\frac{19}{24}$
<newline> </newline>
<newline> </newline>
$^\dagger$: two datasets were single-sex, so a sex effect could not be estimated within-dataset
---
### ComBat does not disrupt within-individual signal
- are we preserving basic properties of the connectome?
<center><img src="images/causal_fmri/signal.png" style="height: 250px"/></center>
---
# Summary
- provide a precise description of batch effects as a causal inference problem
- propose a method for estimating batch effects
- show the batch effect can be removed
- show batch effect removal via ComBat preserves inter-individual and intra-individual signal
---
# Acknowledgements
- Key Contributors: Joshua T. Vogelstein, Brian Caffo, Mike Powell, Noam Finkelstein, Anton Alyakin, Consortium for Reliability and Reproducibility
</textarea>
<!-- <script src="https://gnab.github.io/remark/downloads/remark-latest.min.js"></script> -->
<!-- <script src="remark-latest.min.js"></script> -->
<script src="remark-latest.min.js"></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/KaTeX/0.5.1/katex.min.js"></script>
<script src="https://cdnjs.cloudflare.com/ajax/libs/KaTeX/0.5.1/contrib/auto-render.min.js"></script>
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/KaTeX/0.5.1/katex.min.css">
<script type="text/javascript">
var options = {};
var renderMath = function () {
renderMathInElement(document.body);
// or if you want to use $...$ for math,
renderMathInElement(document.body, {
delimiters: [ // mind the order of delimiters(!?)
{ left: "$$", right: "$$", display: true },
{ left: "$", right: "$", display: false },
{ left: "\\[", right: "\\]", display: true },
{ left: "\\(", right: "\\)", display: false },
]
});
}
remark.macros.scale = function (percentage) {
var url = this;
return '<img src="' + url + '" style="width: ' + percentage + '" />';
};
// var slideshow = remark.create({
// Set the slideshow display ratio
// Default: '4:3'
// Alternatives: '16:9', ...
// {
// ratio: '16:9',
// });
var slideshow = remark.create(options, renderMath);
</script>
</body>
</html>