WFCCM Pre-Post Guide
This is a guide for running pre-post analysis in WFCCM.
For a pre-post analysis, there needs to be a pre-treatment and post-treatment sample for each patient. Missing the pre or post sample will result in that patient being omitted.
NOTE: This is a possibility for a new feature in WFCCM (similar to Average by patient).
Combining data
Until
General Wfccm Design incorporates this process, you'll have to use your favorite spreadsheet. First, all patients need a unique identifier. It helps to specify pre/post by a suffix. For example, the columns
p1-pre
and
p1-post
would represent the pre and post-treatment samples for
p1
. As good measure, I would call the pre-post column
p1
.
Generally, percent change is calculated by
post/pre - 1
. Given the nature of
MicroArray and
MassSpec data, there will be cases when this formula will not suffice.
|
post = 0 |
post <> 0 |
pre = 0 |
|
|
pre <> 0 |
|
|
Log N
Calculating percent change after the log transformation will give different results than before (especially when pre = post). For that reason, we want to calculate percent change on the log-transformed data. Normalized data will suffice in cases such as using Shuo's wavelet preprocessing. In this case, percent change is
log(post)/log(pre)
.
This percent change data will be used for the following methods:
- TTest
- Wilcoxon
- KS
- SAM
- WGA
- InfoScore
- HuWright
|
post = 0 |
post <> 0 |
pre = 0 |
. (NaN) |
? |
pre <> 0 |
? |
log(post)/log(pre) |
Binary
With zero (or missing) values, the percent change does not make sense anymore. Instead, calculate a binary value to run Fisher.
|
post = 0 |
post <> 0 |
pre = 0 |
. (NaN) |
1 |
pre <> 0 |
0 |
? |
Scores and Distance
Run the appropriates scores for the log and binary datasets, then merge. Choose your model and run distance as normal with the log dataset.