0. Experiment Design

Two basic experiment designs:

Design	Description
C-T1T2T3	Compare multiple Test conditions vs one Control
C1T1–C2T2	Compare 2 model systems, each with its own Test and Control

Number of replicates needed:

• Cell lines: minimum 3 replicates / condition
• Primary cell culture: minimum 6 replicates / condition

If the differences in conditions (e.g. treatments) is expected to be low, probably 3 replicates are not enough to show statistically significant differences.

Use biological replicates rather than technical replicates to capture true biological differences between conditions.

There is an option to choose to pool sample lysates, when high biological variability is expected to mask differences between test and control conditions.

---

Balanced Design

Batch effects are variations caused by technical variability due to time, place, and materials (e.g. runs, PamChips). Batch effects can confound real effects if the design is unbalanced which leads to false results.

For example, if all "Test" are in a different run or PamChip than "Control" samples, Test and Control cannot be compared, because there is no way to tell, which part of the difference is technical and which is biological.

A balanced design avoids confounding the effects of the covariates of interest with batch effects. Additionally, batch effects, when they exist, can also be corrected.

Reference: [Leek et al., 2010]

---

How comparisons answer research questions

Assume:

• two cell lines: a wild type (WT) and a knock-out (KO).
• Treatment (T) and untreated (DMSO-treated) Control (C).

Possible research questions and their answers:

1. What is the test effect in WT? → T vs C in WT
2. What is the test effect in KO? → T vs C in KO
3. How is KO and WT different? → KO vs WT in Control (= KO-Control vs WT-Control)
4. How is test effect in KO and WT different?
   1. → simple approach: compare results from questions 1 and 2 with Venn-diagram
   2. → better: compare the fold changes of specific kinases with a low specificity threshold.

Do not use KO-Test vs WT-Test comparison. The reason: KO-Test vs WT-Test does not isolate treatment-response differences because it mixes:

• baseline genotype differences
• treatment-induced genotype differences

The above comparisons are usually done on kinase-level data, the main output of PamDx experiments. Upstream Kinase Analysis is not a statistical test, so interaction effects cannot be calculated. That is why research question 4 is addressed in a simple way.

---

How to Make Sample Annotation

Main sample annotation factors:

Factor	Description
Test condition	Describes the perturbation; what to compare (e.g. treated vs control).
Supergroup	The group within which the test conditions are compared. E.g. WT or KO cell or different tissues.
Sample name	Should be unique to the sample.
Biological Replicate (BR)	The biological subject, e.g. patient, mouse. In case of cell line, cells from different wells in a plate.
Technical Replicate (TR)	Replicates of one biological subject.
PatientID	If experiment has patients, for clarity, include the id.

Example

Barcode	Row	Supergroup ID	Supergroup	Test condition ID	Test condition	Sample name	BR
chip1	1	Sgroup1	Sgroup1_WT	Control	Control	Sgroup1_Control_01	BR01
chip1	2	Sgroup1	Sgroup1_WT	Test	Test_RAFi	Sgroup1_Test_01	BR01
chip1	3	Sgroup2	Sgroup2_KO	Control	Control	Sgroup2_Control_01	BR01
chip1	4	Sgroup2	Sgroup2_KO	Test	Test_RAFi	Sgroup2_Test_01	BR01
chip2	1	Sgroup1	Sgroup1_WT	Control	Control	Sgroup1_Control_01	BR02

---

Naming Rules

• Supergroup ID column values should always be: Sgroup1, Sgroup2, ...
• Test condition ID column values should always be: Control, Test, T1, T2, ...
• Real names go in the Supergroup, Test condition columns
• Have a Supergroup column even if there is only 1 supergroup (for clarity)
• Keep names simple so they are visible on heatmaps.