III. Basic processing

We have run the image analysis workflow. Its output file contains: - Median_SigmBg per peptide-barcode-row-cycle-exposure time - Peptide annotation (Sequence, SwissProtID, gene name, sequence similarity)

Next, add the PTK_template or STK_template to do the data analysis using the output file of the image analysis.

Basic Processing Overview

The result of Image Analysis contains the Median signal minus background (Median_SigmBg) value for each peptide-barcode-row-cycle-exposure time datapoint.

1. data step: add image analysis output files (for PTK: add both prewash and afterwash at the same time)
2. Join with sample annotation
3. Run ETS (Exposure Time Scaling) app. ETS combines Median_SigmBg at multiple exposure times to create a single value per cycle: S100 (AU).
4. Check Mean and CV (CV is the data quality metric)
5. Check positive control (pre-phosphorylated peptides). Expectation: S100 > 1000
6. Run the QC app (PTK or STK QC app)

---

ETS: Exposure Time Scaling App

The ETS is the first app in PamDx data analysis. An app is a subworkflow: left click "Open" to inspect the structure to familiarize yourself with how an app looks like in general. It always starts with an Input step, which takes the data from the main workflow. Then there is a wizard step which helps the users to select data. Then there are data transformation steps and finally: - View step(s) to visualize data - Output step to bring data back to the main workflow

However, usually we do not have to interact with the app from inside. When we run the app from the main workflow, the wizard pops up, where we have to select correct data input by answering questions.

In ETS, the correct factors are set by default — just click Run.

Sample quality control: the Mean & CV operator

We examine the sample quality by assessing the variance in the individual conditions / samples. Each condition has usually 3+ biological or techncial replicates. An example of technical replicate is 1 sample lysate on 3+ arrays; An example of biological replicate is 1 sample lysate per array).

How to configure the crosstab view - Each column should contain all replicates of a condition. - Colors should be set per condition (e.g. drag and drop Supergroup and/or Test condition to colors)

Running the operator switches to operator view; you can switch back to crosstab view.

- If CV for high signals is NA → adjust the curve fit parameters shown in the panel - Result: CV high signals per condition. To interpret results (whether the variation is high or low), use the quality control criteria table below.

---

QC Apps

QC apps calculate which peptides have low signal. Phosphorylation on these peptides is not detected, and these peptides should be removed. Peptide removal does not happen inside the app - it will be removed at a later step.

PTK QC app

1. Calculates presence value per peptide-array: positive if peptide signal shows positive trend over cycles.
2. Calculates fractionPresent: the fraction of arrays with a positive trend per peptide.
3. At a later step, the fractionPresent factor is used to remove peptides which are only present in a fraction of all the samples (default: 25% which can be lowered.)

Run the app and view the visualization to assess peptide presence:

STK QC LOD app

• This app calculates the same factors as the PTK QC, but in a different way. The reason is that the prewash steps cannot be used in the STK assay to assess the trend over the cycles.
• Presence is calculated based on the Limit of Detection (LOD). Peptides with S100 above the LOD are considered present.
• Per peptide, the ratio of samples that are present is calculated in the qc_stk_thr.ratio factor.

Next, at the first normalization, the peptides which are present in < 25% of samples are excluded. 25% is the default threshold but it should be always checked and adjusted. 25% stems from the following rational: If there are 4 conditions, 25% = 1 condition. It is usually acceptable if peptides are missing in max 1 condition.

Always review this threshold during analysis! Sometimes, we can accept lower threshold, e.g. if there are not enough peptides passing QC or with potent inhibitors which inhibit kinases so that signal is only present in the control sample.

STK QC App: Use When All Peptides Pass the LOD App

If pre-phosphorylated peptides have very low signal, all peptides will pass the STK_QC_LOD app. In this case, use the STK_QC app (based on CV).

• Calculates nominal CV
• At normalization, remove peptides with high CV (= low signal). Use the pep.cv.estimate factor. The default threshold is < 0.5, but revise this.

Removal of low signal peptides

Removal of low signal peptides is done by a Filter in the log step or, if VSN normalization is done, then in the Filter QC peptides step after VSN. These filters are set up with the defaults in the templates. The QC heatmaps should be used to visually assess if these thresholds should be adjusted.

Quality Control criteria table

The following criteria can be used to assess quality of the experiment. The quality control does not serve to exclude samples - it rather manages our expectations. For example, if the number of QC passed phosphosites is high and the samples have low biological CV, we can expect that there is a differential signal between conditions. On the contrary, if the flags indicate fair or poor sample quality, probably differential effects won't be highly significant .

Level	Flag	Number of QC-passed phosphosites PTK (of 195)	Number of QC-passed phosphosites STK (of 142)	Technical CV	Biological CV
Good	🟢	> 123	> 90	< 20%	< 30%
Fair	🟡	78–123	56–90	20–30%	30–40%
Poor	🔴	< 78	< 56	> 30%	> 40%

---

Data flow

In general the data flows in the following way. An operator transforms the data (e.g. log transform). It creates an output factor. Each step has a namespace which is the name of the step.

Output construction follows the pattern: <namespace>.<output_name_from_operator>

In the next step, the output of the previous step is taken as input values. It is represented on the y axis of the crosstab view.

Example: - Namespace: STKQC - Output name: logTransformed - → Full reference: STKQC.logTransformed

---

Normalizations

Read about different base normalizations here.

Normalizations in the Tercen workflow

Normalization is done together with filtering

Log normalization & filtering for QC-passed peptides is done in 1 step.

VSN normalization & filtering for QC-passed peptides is done in two steps. The reason is that log works per individual value, while VSN considers all peptides simultaneously - even peptides not passing QC are needed to carry it out.

Filtering for peptides that passed QC is done with the following defaults:

• PTK: fractionPresent > 0.249
• STK: qc_stk_thr.ratio > 0.25

The log_cutoff operator assignes 1 to values < 1 before log transformation.

---

ComBat: Batch Effect Correction

1. Crosstab view: color by expected technical batch (Barcode for single PS12 run; Run for multi-run projects)
2. Operator view: assess batch effect removal — check if before-ComBat PCA shows batch clusters and after-ComBat PCA has resolved them
3. If resolved, click Done — leave the page only when the dot appears under Done

Check PCA and Heatmap Before and After ComBat

If ComBat successfully decreases technical variance, then after Combat:

• Replicates within a condition are more similar &
• Conditions are usually more separated

Important: for multi-run experiments: check Run effect first (usually larger than Barcode). Correcting for too many batch types can introduce bias — if Run effect exists, correcting for Barcode as well is usually unnecessary.

---

References

ComBat and batch correction:

• Chen, C., et al. (2011). Removing batch effects in analysis of expression microarray data: An evaluation of six batch adjustment methods. PLoS One, 6, e17238.

• Johnson, W.E., Li, C. & Rabinovic, A. (2007). Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics, 8, 118–127.

• Leek, J.T., et al. (2010). Tackling the widespread and critical impact of batch effects in high-throughput data. Nature Reviews Genetics, 11(10), 733–739.

• Zindler, T., et al. (2020). Simulating ComBat: how batch correction can lead to the systematic introduction of false positive results in DNA methylation microarray studies. BMC Bioinformatics, 21(1), 1–15.