Protein-protein interaction analysis: scoring AP-MS and BioID, building networks, modeling structures
Analyzing protein-protein interaction data: SAINT and CRAPome scoring for AP-MS and BioID, quantitative interactomics, networks, AlphaFold-Multimer.
Key takeaways
- Most proteins in a pull-down are background. Score interactions against matched controls (SAINTexpress) and contaminant repositories (CRAPome).
- Proximity labeling (BioID, TurboID) maps neighborhoods of roughly 10 nm, not only direct binders.
- Database networks (STRING, BioGRID) add context but carry literature bias toward well-studied proteins.
- AlphaFold-Multimer predictions are hypotheses; report interface confidence (ipTM, pDockQ).
Protein-protein interactions explain function: which complex a protein joins, which pathway it signals through, which partner a mutation disrupts. Experimental interaction data is noisy by nature. Every purification pulls down sticky, abundant proteins along with real partners. The analysis is mostly about separating the two.
Experimental approaches and what they measure
| Method | What it captures | Main analysis challenge |
|---|---|---|
| AP-MS / co-IP-MS | Stable complexes with the bait | Nonspecific binders |
| BioID / TurboID / APEX | Proteins within ~10 nm over time | Neighbors vs direct binders |
| Cross-linking MS | Residue-level contacts | Search complexity, FDR |
| Yeast two-hybrid | Binary interactions | False positives and negatives |
| Co-fractionation MS | Complexes across fractions | Profile correlation scoring |
Step 1: Quantification
Search raw files with MaxQuant, FragPipe or DIA-NN, with protein-level FDR control. Choose quantification (spectral counts, MS1 intensity, or TMT for multiplexed comparisons) to fit the scoring method, and keep all controls in the same search.
Step 2: Score against controls
- SAINTexpress models each prey's abundance in bait purifications against negative controls (empty vector, GFP-only bait, IgG) and returns a probability and Bayesian FDR.
- Fold change over control adds an effect-size view.
- CRAPome frequencies show how often a protein appears in unrelated pull-downs. Frequent fliers such as ribosomal proteins, chaperones and keratins need very strong evidence.
What our experts avoid: calling every protein enriched over a single control an interactor. We require replicate controls, probabilistic scoring and background frequencies. A bait-specific network of 30 confident interactors is more useful than 300 enriched proteins.
Step 3: Quantitative interactomics
When the question is how interactions change with mutation, drug treatment or stimulation, use quantitative designs (label-free with replicates, or TMT) and test differential interaction with limma-style models. Normalize carefully to bait abundance, because a bait that is expressed less pulls down less of everything.
Step 4: Networks
- Combine experimental interactors with curated databases (STRING, BioGRID, IntAct) for context.
- Detect complexes and modules with clustering (MCL, ClusterONE) and test enrichment against known complexes (CORUM).
- Network propagation prioritizes genes from screens or GWAS by their proximity to disease modules.
- Remember literature bias: well-studied proteins have more known interactions, which inflates their apparent centrality.
Step 5: Structural modeling
AlphaFold-Multimer and AlphaFold 3 can model candidate binary interactions and complexes. We use interface confidence (ipTM, pDockQ) to rank candidates, map disease variants and PTMs onto predicted interfaces, and propose mutations for validation. Low-confidence interfaces stay hypotheses.
How OmicsQuest can help
Our PPI analysis service scores AP-MS, proximity labeling and Y2H data, builds Cytoscape-ready networks and models complexes. For global and phospho-proteomics, see proteomics analysis.
Frequently asked questions
What is SAINT scoring?
SAINT (Significance Analysis of INTeractome) estimates the probability that a protein detected with a bait is a true interactor, by comparing its spectral counts or intensities with negative control purifications. SAINTexpress is the fast, widely used implementation. A Bayesian FDR threshold (commonly 1-5%) defines high-confidence interactors.
Can AlphaFold predict protein-protein interactions?
AlphaFold-Multimer and AlphaFold 3 can predict structures of protein complexes, and their confidence metrics (ipTM, pDockQ) help rank candidate interactions. They work well as a filter and to propose interfaces, but predictions need experimental validation, especially for transient or disordered interactions.
Work with OmicsQuest. Related services:
References
- Teo G, Liu G, Zhang J, et al. SAINTexpress: improvements and additional features in Significance Analysis of INTeractome software. J Proteomics. 2014;100:37-43.
- Mellacheruvu D, Wright Z, Couzens AL, et al. The CRAPome: a contaminant repository for affinity purification-mass spectrometry data. Nat Methods. 2013;10:730-736.
- Szklarczyk D, Kirsch R, Koutrouli M, et al. The STRING database in 2023: protein-protein association networks and functional enrichment analyses for any sequenced genome of interest. Nucleic Acids Res. 2023;51:D638-D646.
- Jumper J, Evans R, Pritzel A, et al. Highly accurate protein structure prediction with AlphaFold. Nature. 2021;596:583-589.
- Cox J, Mann M. MaxQuant enables high peptide identification rates, individualized p.p.b.-range mass accuracies and proteome-wide protein quantification. Nat Biotechnol. 2008;26:1367-1372.
- Demichev V, Messner CB, Vernardis SI, Lilley KS, Ralser M. DIA-NN: neural networks and interference correction enable deep proteome coverage in high throughput. Nat Methods. 2020;17:41-44.