Skip to content

4. Further insights to PamDx data

The interpretation of peptides and kinases: which one is more insightful ?

We suggest to base further analysis of PamDx results on kinase data and not on peptide-level data.

  • Peptide signal can be translated to protein signal (data provided in peptide enrichment) but interpreting peptides = proteins is oversimplification:
    • the peptide signal is the sum of phosphorylation by different kinases on a 13mer peptide on the PamChip.
    • The protein may not be present in lysate.
    • Phosphorylation can have different outcomes (induction or inhibition of protein or other functions).
  • The UKA results on the other hand, are directly suitable to input to signaling pathway or network analysis:
    • they have a clear interpretation: positive FC = activation, negative = inhibition.
    • Differential kinase activity has a better statistical underpinning than treating differential peptide phosphorylation as protein activity. (Specificity score is available for kinases and kinases can only be predicted if at least 3 phosphosites show consistent change.
    • Since kinases are predicted, pre-selection of kinases with high specificity scores (> 0.7, or > 1, the latter is equivalent to p-value 0.1) should be done. We do not recommend to set stringent specificity score cutoff for pathway / network analysis. These analyses find functionally interacting kinases, which further pinpoints the most relevant ones and naturally filters out false positives. (If a kinase is not connected to any other kinase in a pathway or network analysis, it is likely to be false positive.)
    • since PTKs and STKs usually act together in pathways, combine them for pathway and network analysis.

I really can't use peptide signal?

You can decide to use peptide signal as a proxy for protein activity in certain cases. For example, some peptides may match to proteins which are not kinases and they are specifically linked to the research domain. Or, if you want to use autophosphorylation site which is a strong indicator that the kinase is there (although in this case the kinase should be predicted as well.) If you do use peptides translated to proteins, mention the simplifications in your publication.

Example of phosphosites on the same protein with different functions:

ID PSite Residue Function
AKT1_301_313 308 T required for full activity
AKT1_170_182 176 Y phosphorylation by TNK2 tethers AKT1 to plasma membrane

Pathway and Network Analysis Are Complementary

Pathway-network analysis

Pathway Analysis Tools

Type of analysis:

Choose the pathway databases, e.g:

  • KEGG (nice pathway figures but limited gene sets)
  • Reactome (most recommended by us because of clear hiearchical structure and constant update)
  • WikiPathways (nice pathway figures, but inconsistent pathway naming. Some diseases are only found here.)
  • GO
  • SignaLink very well curated db
  • Hallmarks of Cancer

Pathway analysis types

Overrepresentation (e.g. Metascape) Functional class scoring (e.g. GSEA)
TLDR: More true positive, more false positive results.

Tests whether the overlap of pathway members in the data is higher than by chance.

Input: list of significant kinases (no fold change or p-value needed)		 TLDR: Less false positive, less true positive results.

Detects pathways based on subtle, coordinated changes in pathway members.

Input: fold change of kinases.
Pro: easy to use. Pro: Can sensitively detect subtle, coordinated changes in pathway members. Pathway output is more reliable.
Con: Pre-filtering of features can introduce bias: 1) subtler changes might not be significant; 2) multiple testing bias (inflation of false positives) Con: Can overlook pathways if members don’t show clear up-or downregulation.

Additional interesting tools

Lomics is an LLM-based tool that gives potential pathways and pathway members upon a question.

Protein Atlas is useful to verify biological context of a kinase hit. (E.g.: is LCK expressed in endothelial cells?)

Dark Kinome provides information about not-well-studied kinases (network, expression, pathways, assays to study).

To overlap the top hits across different comparisons (e.g. T1 vs C, T2 vs C, etc.) you can use Venny.

Example outputs:

Paintomics output: KEGG pathways colored by LFC (of 1+ comparisons)

Paintomics output

Metascape output: top pathways:

Metascape output

Hallmarks of cancer

Pathway-network analysis|500

How to select the most relevant pathways?

There is no simple metric like a p-value to select the most relevant pathways. Pathways share members, and a pathway that is relevant in one model system may be irrelevant in another.
In addition, statistical pathway enrichment tests do not consider how specific the found members are to a pathway. As a result, a pathway may receive a strong p-value even if only non-specific members are detected. Therefore, we recommend assessing pathway relevance using biological domain knowledge rather than relying on p-values or combined scores.

Network analysis tools

To understand exact interactions between kinases and cross-talks between pathways, use network analysis.

SIGNOR STRING db
Features Experimental evidence about causal interactions.

Pathways (signaling, metabolic, disease)		 Network from experimental sources and association (e.g. Gene-cooccurence, text-mining) sources
Direction Directed Mostly undirected
Mechanistic detail High Low–moderate
Best for Inferring signaling pathways and causal networks Finding interacting proteins and network modules
Weakness Smaller coverage Many edges are not causal or signaling-specific

SIGNOR

Use the multi-protein search tool with connect mode to find a network of the top kinase list. The resulting network might include connector nodes (connectors between kinases without direct link).

STRING

Use the Search multiple proteins module and paste your top kinase list to find how they are connected.

To focus on the most validated part of the network, filter settings to make the interaction data more stringent. See interaction sources / evidence channels in the below table. You can also click on a link to see the source of interaction with publications.

Evidence Channel Type Reliability Key Strength
Experimental Direct 🥇 Highest Proof of direct physical or genetic interaction.
Curated Databases Direct 🥇 Very High Expert-validated functional association in pathways/complexes.
Gene Fusion Genomic 🥈 High Strong evolutionary evidence for functional interaction.
Gene Neighborhood Genomic 🥈 High (Prok) / Low (Euk) Evidence for operons and co-regulation (mainly in bacteria).
Gene Co-occurrence Genomic 🥈 Medium-High Evidence for participation in the same pathway/system.
Co-expression Functional 🥈 Medium Strong evidence for co-regulation and functional association.
Text mining Literature 🥉 Lower High-level associative

Network clustering also helps to understand the network. The clustering methods require an initial guess of how many clusters you expect - you can play with this setting. The clustered networks can help to see pathways.

Suggestions for validating results with Western blot

Select kinases that appear with high Specificity Score in experiments with ≥ 3 replicates/condition (cell lines) or ≥ 6 replicates/condition (primary cell cultures / patient samples).

Western blot validation strategy:

  • Upregulated kinase: use antibodies specific for phosphosites that induce kinase activity

  • Downregulated kinase: use antibodies specific for phosphosites that inhibit kinase activity

Use Uniprot and PhosphoSitePlus to identify relevant phosphosites. PamGene can also help identify relevant phosphosites on request.

If direct kinase validation is not possible, validate downstream phosphosites activated by the kinase (PamDx can assist on request).

Questions?

For questions about experiment design, data analysis, validation, or anything else: support@pamdx.com