Experimental Design May 2007 Hanne Jarmer. What is Experimental Design? Question Answer Data...

Experimental DesignMay 2007

Hanne Jarmer

What is Experimental Design?

Question

Answer

Data Analysis

Experiments

Often not the final answer!

What do we need?

• What is the question(s)?

• Which experiment(s) will give the answer?

• How many replicates do we need?

The “correct” experimental design?!

• Several equally good designs

• Balancing

• Replicating

The Question

What is a good question?

... that can be answered by the use microarray experiments?

The Questions

Why do obese children have diabetes more frequently?

What is the connection to fitness/diet?

?

The Experiments

1. From question: Define your categories

2. Access to material? (living human brain?)

3. How many replications?

The Experiments

Determine from where you’d get variation?

Hopefully: Biology !!

- array

- spot

- time

- patient gender/age/other

- dye

- technician

The Experiments

From question: Define your categories

Sick < > healthy

The Experiments

Access to material? ... Mmmm, three of these ...

Human samples not always possible ...

Maybe rats instead?

The Experiments

How many replications?

Impossible to answer!

Depends on how ...

... strong the biological signal is

... much you’d like to be able to trust the result

... you plan to analyze

... much you can afford ...

Biological replications are always preferable!

The Experiments

Variation ...

How do we get (only) what we want?

THE KEY: BALANCING .. all other sources of variation

SICK MEN HEALTY MEN

2 old, 6 young 2 old, 6 young

3 day A, 3 day B 3 day A, 3 day B

2 batch 1, 4 batch 2 2 batch 1, 4 batch 2

Technician Y Technician Y

DYE SWAP ... is it really necessary?

The non-linear normalization should take care of this ... right?!

This is mostly true,... but not always:

“Technical” or “biological” dye swap?

A biological replication always give you more !!

The Experiments

Important: Pick conditions that:

... will maximize the interesting biological signal

Ex.: Wild-type bacteria < > TF-mutant

... and ...

choose a growth medium where the TF is active!!

TF

List of genes?

The Experiments

Important: Pick conditions that:

... will maximize the interesting biological signal

... while minimizing the overall difference

Less optimal: Liver < > Brain

Basic Designs

“Reference design” “Loop design”

Stanford-type microarrays:

Basic Designs

“Loop design”

2 samples from the same category

Please, take home...

1. Question How to save the world?

2. Categories A - B -C

3. Balance

4. Replication (biological)

Example: Why not AIDS?

1. Question: Why do some HIV-infected ... not develop AIDS?

2. Categories:H

IV-i

nfec

ted

Patient type

0100

10 11

0 = no AIDS

1 = AIDS

0 = no HIV

1 = HIV

NEVER AIDS

Example: Why not AIDS?

Time

Blood samples

Blood samples

HIV

AIDS

00

01

10

11

~ 20 years

HIV

-inf

ecte

d

Patient type

0100

10 11

0 = no AIDS

1 = AIDS

0 = no HIV

1 = HIV

Determination of categories

Example: Why not AIDS?

Balancing and replicating

Variation can come from:

Village, age, doctor, .... TH-cell extraction

Relatively few “0” patients ...

- Pick corresponding “1” patients

Example: Why not AIDS?

HIV

-inf

ecte

d

Patient type0 = no AIDS

1 = AIDS

0 = no HIV

1 = HIV

01

11

00

10

3 P-values for each gene ...

Example: Why not AIDS?

The 2-way ANOVA results:

3 lists:

- Patient-type difference

- +/- HIV difference

- The Interaction:

“0”-patient genes that behave

unexpectedly during HIV-infection

Example: Why not AIDS?

- CCR5-mutation (no HIV)

- The golden combination of HLAs (HLA-B57 in particular)

- Something unknown !?

Please, take home...

1. Question How to save the world?

2. Categories A - B -C

3. Balance

4. Replication (biological)

Short Discussion

A) Questions:

1) How will Treatment A or B or the combination of both affect the genes in the malaria parasite?

2) Will either affect the production of the “sticky” proteins responsible for the aggregation of the red blood cells?

Short Discussion

B) How would you design this study?

C) Which categories?No treatment

Treatment A

Treatment B

The combination

Short Discussion - in groups

The study is done using two-color microarrays and 4 biological replications

D) Which samples would you label red and which would you label green?

E) What will you hybridize with what?

N-1 A-1 B-1 AB-1

N-2 A-2 B-2 AB-2

N-3 A-3 B-3 AB-3

N-4 A-4 B-4 AB-4

N-1 A-1 B-1 AB-1

N-2 A-2 B-2 AB-2

N-3 A-3 B-3 AB-3

N-4 A-4 B-4 AB-4

Short Discussion - in groups

N-1

B-1

A-2

AB-2

N-3

B-3

A-4

AB-4

N-2

AB-3

A-3

B-4

AB-1

N-4

B-2

A-1

Short Discussion - in groups

A) Can you come up with another interesting biological question?

B) Which categories would you need?

C) How would you balance this study?