en-fr
en-sl
en-es
en
0.25
0.5
0.75
1.25
1.5
1.75
2
Removing unwanted variation in machine learning for personalized medicine
Published on 2016-07-181230 Views
Machine Learning for Personalized Medicine will inevitably build on large omics datasets. These are often collected over months or years, and sometimes involve multiple labs. Unwanted variation (UV) c
Related categories
Presentation
Removing Unwanted Variation in Machine Learning for Personalized Medicine00:00
Apology, Motivation and Declaration of Conflict of Interest02:27
Afirma - 103:15
Afirma - 205:21
Introduction to our RUV methods05:42
The problem06:03
Artifact can overwhelm biology06:19
Some scientific goals sought using gene expression microarrays07:59
Aim for today08:15
“Our” model (brief refs later)08:36
Concrete example13:37
Our model in pictures14:48
Our goal: classification16:40
Our model, 216:57
Some ways of dealing with these and related problems with microarrays17:21
Identifiability: we don’t know the correlation of W (k=1) with X19:13
We might have genes j not affected by X - 120:34
We might have genes j not affected by X - 221:06
“Our” solution: Use control genes22:45
Using the negative controls c24:42
Introducing the two-step: RUV-226:51
Removing severe batch effects32:10
Standard analysis32:31
A microarray experiment with central retina tissue from the rd1 mouse35:03
Analysis with RUVinv35:25
Are there any questions?41:53
Introducing RUV-inv44:29
Classification44:53
What is unwanted variation? - 147:30
What is unwanted variation? - 249:36
Hypothetical example49:53
The challenge of non-stationarity54:06
An interesting point54:27
Removing unwanted variation from the test (target) set55:26
Model for the training set data55:57
Model for the test and target set56:20
Goal57:52
How to proceed?58:27
Method A start with αˆ01:00:43
Digression: some calculations01:02:15
Method B, start with ˆβ01:03:16
Comparing and contrasting Methods A and Method B01:04:18
Choice of classifier01:06:34
Advantage of P(B)01:10:47
Cleaning up the training set01:11:02
Cross Normalization01:16:01
How does it work? Simulations01:16:19
Removing Unwanted Variation makes it possible to use “simple” classifiers01:16:24
Why we might be able to stick to “simple” classifiers?01:16:39
Example01:16:41
Gender differences in the brain01:16:46
Ex: gender differences in the brain01:17:44
Principal component01:17:54
Same plot with gender indicated01:18:59
Same plot with brain regions indicated01:19:20
Ex: gender differences in the brain, 201:19:35
Estimated accuracy rates01:21:36
Making this work for personalized medicine01:26:26
Here are a couple of thoughts01:26:29