Biology Atlas

Map / Assays

assay · intermediate · partly written

Affinity purification mass spectrometry (AP-MS)

AP-MS hooks one tagged protein out of broken-open cells, washes, and uses a mass spectrometer to count which other proteins came along, giving a statistical statement of co-complex membership rather than of direct binding.

Also called: AP-MS, IP-MS, affinity enrichment MS, AE-MS, pull-down mass spectrometry

What it is

AP-MS captures one tagged protein from lysed cells with its partners, then identifies what stayed attached with a mass spectrometer. A score compares each protein's count with background. It tells you who sat in the same assembly in that cell type under that wash, not who touches whom. 12

Reported, not yet replicated

From peptide counts to a probability more

A fixed cut-off treats a sticky abundant protein and a rare real partner alike, so scoring models estimate what counts the controls would produce. SAINT assigns each pair a posterior probability of a true interaction, averaged over replicates; its authors say FDR accuracy remains to be validated. On one 27-bait dataset with 35 controls a probability of 0.9 kept 1,375 interactions. SAINTexpress at a probability of at least 0.8 gave 639 versus 697 interactions for two versions of the software, 584 shared. In the CRAPome paper a fold change of 4 is an example of a high-scoring interaction, not a cut-off. 234

Reported, not yet replicated

What the washes decide more

The washes are the sieve: strong or slow-to-leave partners survive and weak or fast-exchanging ones are rinsed away, which is where false negatives come from. The BioPlex authors attribute their false negatives to the stringency of biochemical purification and statistical filtering. The same logic limits pull-down experiments in general; see the co-immunoprecipitation record. 15

Reported, not yet replicated

Scale and cost deep

BioPlex 3.0 reports 10,128 baits and 118,162 interactions among 14,586 proteins in 293T cells. At one university core, pulldown plus measurement is priced at $850 per sample internally and $1,360 externally (rate card effective 2025-11-01), before antibody, analysis and your own time. A 12-sample design at the external rate is about $16,300 plus analysis on our arithmetic, not a quote. 16

Reported, not yet replicated

Question answered

Which proteins stay with this bait, in this cell type and under this wash, in amounts above background? A hit is evidence of co-complex membership, not of contact. 12

Reported, not yet replicated

Readout physical

Peptide-spectrum matches. The captured proteins are chopped into peptides by trypsin, separated by liquid chromatography and fed into a mass spectrometer, which fragments each peptide and matches the fragment pattern to a sequence database. The number of matches assigned to a protein is its spectral count, used as a coarse proxy for how much of it was present. 12

Reported, not yet replicated

Not measured

Contact: in BioPlex 3.0, only 32% of edges that fall within 309 structures were direct contacts. Weak or fast-exchanging partners are lost in the washes, and the BioPlex authors write that "false negatives may prevail". An AP-MS score is a property of the bait and its background, so the same prey can be a hit for one bait and noise for another. No kinetics, no affinity. 12

Reported, not yet replicated

Setting

Where in cascade

Sample prep

A bait gene with an epitope tag (HA, FLAG or GFP) is expressed in cells. Cells are broken open in detergent buffer and antibody beads against the tag capture the bait and what is attached. BioPlex 3.0 protocol example: lysis in 50 mM Tris pH 7.5, 300 mM NaCl, 0.5% NP40; anti-HA beads for 4 hours at 4 C; four washes in lysis buffer and two in PBS; elution with HA peptide; trypsin digestion; each sample run twice by LC-MS of about 70 minutes each. 1

Reported, not yet replicated

Conditions

The result depends on the wash stringency, the tag, the resin and the cell type. In the CRAPome repository, 14 proteins were found in more than 90% of 360 control experiments and 89 in more than 50%, whereas 3,571 proteins (80%) were seen in 10% or fewer: "contaminant" is a property of a protein plus a protocol. SAINT is advised with at least 3-5 matched control purifications, ideally; the CRAPome pools controls from 12 labs to supply them. BioPlex baits were tagged ORFeome clones expressed in 293T cells and were not at endogenous level. 421

Reported, not yet replicated

Throughput

BioPlex 3.0 analysed 10,128 baits in 293T cells and reported 118,162 interactions among 14,586 proteins, which is about half of the proteome in one cell line. Each sample took two LC-MS runs of about 70 minutes. We set the class to medium as an estimate; the paper gives no per-day rate. 1

Our inference

Time per run

Capture of 4 hours at 4 C plus two LC-MS runs of about 70 minutes each in the BioPlex protocol (sample preparation and data analysis not included). A different, flow-through AE-MS protocol took about 2.5 hours from lysis to elution, with 15 minutes of capture at 4 C. 15

Reported, not yet replicated

Cost

Per sample

850 to 1,360 USD per sample, pulldown plus measurement (Baylor College of Medicine Mass Spectrometry Proteomics Core rate card, "Protein Interactome Analysis": pulldown $350 + measurement $500 per sample at the BCM internal rate, $560 + $800 at the external academic rate. Antibody not included; data analysis is extra ($700-1,120 per 12 samples). An academic core-facility recharge rate, not a commercial or contract-lab price; no commercial price was found.; as of 2025-11-01 (rate card effective date; fetched 2026-11-17))6

Engineerable variants

Choice of tag and resin

The same prey can look like a contaminant with one tag and a real partner with another: in the CRAPome, STK38 co-purified in nearly all FLAG experiments but not GFP, and TP53 mostly in GFP experiments. 4

Reported, not yet replicated

Scoring against controls: SAINT and SAINTexpress

SAINT models each bait-prey pair as a mixture of two hidden states, true or false interaction, each with its own distribution of expected counts, with the false distribution estimated from the control purifications; counts are normalised for protein length and total spectra. SAINTexpress uses a two-component mixture of generalized Poisson distributions, adds a topology score and cut runtime from 37 minutes to 20 seconds on one test. 23

Reported, not yet replicated

Scoring against the pool: CompPASS and CompPASS-Plus

CompPASS needs no controls and scores each prey against all other purifications in the pool; in its founding paper 774 of 2,458 proteins qualified as high-confidence and 45 of 66 tested interactions were validated by an independent method. CompPASS-Plus (used for BioPlex) is a naive Bayes classifier trained per 96-well plate. 71

Reported, not yet replicated

Fold-change scores from pooled controls

CRAPome fold change FC-A divides the prey's normalised count in the bait run by its average across controls; FC-B uses the mean of the highest three control values, which catches contaminants that spike in a few controls. 4

Reported, not yet replicated

Affinity enrichment (AE-MS)

A variant that does not purify to homogeneity and instead compares each tagged strain with the other tagged strains, treating the roughly 2,000 background binders in each pull-down as a normalisation resource. Demonstrated in yeast. Only the first half of the full text was read. 5

Reported, not yet replicated

Endogenous-level tagging

Tagging the endogenous gene to avoid overexpression is described in the literature as an alternative; we did not open a source that quantifies its effect. 5

Unverified: from our own report, source not opened

Use cases

SARS-CoV-2 protein interaction map and drug repurposing (Gordon 2020)

AP-MS of 26 of 29 viral proteins gave 332 high-confidence virus-human interactions, 66 druggable host proteins and 69 compounds; sigma receptor regulators showed antiviral activity. A later paper from the same network found that for 23 cationic amphiphilic drugs, including hydroxychloroquine, antiviral effect tracked drug-induced phospholipidosis, which the abstract says "does not reflect specific target-based activities". Our reading: the interactions were not shown to be wrong; the step that failed was moving from an interaction to a drug mechanism. Both papers were read as abstracts only; the clinical-trial outcome of hydroxychloroquine is not sourced here. 89

Our inference

BioPlex 3.0, an all-against-all style map

10,128 baits in 293T cells gave 118,162 interactions among 14,586 proteins; a second network came from HCT116 cells. The authors state that false negatives may prevail. 1

Reported, not yet replicated

ML role

Does

Statistical scoring: SAINT-family mixture models turn spectral counts into probabilities, and CompPASS-Plus uses a naive Bayes classifier with features such as z-score, NWD score, entropy between technical replicates and detection frequency. Decoy sequences supply "false" labels and previously published interactions supply "true" ones. 231

Reported, not yet replicated

Cannot

The SAINT authors state that the accuracy of the FDR estimate remains to be validated. Training "true" labels drawn from earlier literature risk circularity (our inference). A score cannot say which prey touches the bait. 21

Our inference

Data availability

BioPlex interaction networks public

Downloadable networks from the group (293T 3.0, HCT116 and others). 101

Caveats. The site states no data licence (copyright footer only) and asks users to cite the three BioPlex papers; for the group's unreleased quarterly data it asks users to hold off on large-scale analysis. Scores compare within a cell line; the network is one cell line and co-complex, not binary.

Reported, not yet replicated

CRAPome contaminant repository public

Version 1.0 held 360 negative-control experiments from 12 labs (343 human) for scoring against controls. 4

Caveats. Contaminant frequencies depend on tag, resin and cell type; match controls to your protocol.

Reported, not yet replicated

University mass-spectrometry core (Baylor College of Medicine rate card) purchase

Pulldown plus measurement at about $850 (internal) to $1,360 (external academic) per sample, plus analysis. 6

Caveats. Academic recharge rates that depend on customer type; exclude antibody, bait cloning and your staff time. A contract research organisation price was not found on any provider page.

Price. 850 to 1,360 USD per sample (BCM core rate card, pulldown + measurement; as of 2025-11-01)6

Reported, not yet replicated

Northeastern University Mass Spectrometry core (NUMS) purchase

Instrument time, digestion, TMT labelling and data analysis are priced as separate lines; there is no per-sample AP-MS line. 11

Caveats. Core recharge rate effective 2025-09-01, not a market price; billed by hours booked; building a per-sample cost from the parts is not recorded here.

Price. 75 to 145 USD per instrument hour (academic to industry) (NUMS rate card effective 2025-09-01; digest $25 or $50 per sample; analysis $100 or $170 per hour.; as of 2025-09-01)11

See: Northeastern University Mass Spectrometry core (NUMS)

Reported, not yet replicated

Teach this

In one sentence

AP-MS fishes out one protein from broken cells and counts what comes along, then asks statistically whether each catch is more than the sticky background. 21

Reported, not yet replicated

The five-minute version

Put a tag on the protein you want (the bait), break the cells open gently, and capture the bait with antibody beads. Wash. Whatever is still attached is chopped into peptides and identified by a mass spectrometer, which counts how many times each protein was seen. Many proteins stick to beads or tags regardless, so a scoring method compares each count with control purifications (SAINT, the CRAPome) or with all other purifications in the pool (CompPASS) and returns a probability. The result is "enriched over background in this cell and protocol", which says the proteins sat in one assembly, not that they touch. 241

Reported, not yet replicated

What a listener will ask

Does a high score mean they bind each other directly?

No. In BioPlex 3.0 only 32% of edges within 309 structures were direct contacts. 1

Reported, not yet replicated

Why do controls matter so much?

Some proteins show up in most purifications: 14 proteins appeared in more than 90% of 360 CRAPome control experiments. A count only means something relative to the background. 4

Reported, not yet replicated

What does it miss?

Weak or quickly exchanging partners that wash off; the BioPlex authors write that false negatives may prevail. 1

Reported, not yet replicated

The common wrong idea

A high-probability AP-MS interaction is a measured binding event.

It is a probability that the prey is enriched over background; it supports co-complex membership only.

The one citation to hand over: 1

Still open here

Not yet written: Before after, Capital.

Linked from

Sources

  1. Huttlin, E. L., Bruckner, R. J., et al.. Dual proteome-scale networks reveal cell-specific remodeling of the human interactome. Cell (2021). paper; opened 2026-11-17 all records citing this
  2. Choi H, et al.. SAINT: probabilistic scoring of affinity purification-mass spectrometry data. Nature Methods (2011). paper; opened 2026-11-17 all records citing this
  3. Teo G, et al.. SAINTexpress: improvements and additional features in Significance Analysis of INTeractome software. Journal of Proteomics (2014). paper; opened 2026-11-17 all records citing this
  4. Mellacheruvu, D., et al.. The CRAPome: a contaminant repository for affinity purification-mass spectrometry data. Nature Methods (2013). paper; opened 2026-11-17 all records citing this
  5. Keilhauer EC, Hein MY, Mann M. Accurate protein complex retrieval by affinity enrichment mass spectrometry (AE-MS) rather than affinity purification mass spectrometry (AP-MS). Molecular and Cellular Proteomics (2015). paper; opened 2026-11-17 all records citing this
  6. Baylor College of Medicine Mass Spectrometry Proteomics Core: fees. Baylor College of Medicine. organisation-page, not peer reviewed; opened 2026-11-17 all records citing this
  7. Sowa ME, et al.. Defining the human deubiquitinating enzyme interaction landscape. Cell (2009). paper; opened 2026-11-17 all records citing this
  8. Gordon, D. E., et al.. A SARS-CoV-2 protein interaction map reveals targets for drug repurposing. Nature (2020). paper; opened 2026-11-17 all records citing this
  9. Tummino, T. A., et al.. Drug-induced phospholipidosis confounds drug repurposing for SARS-CoV-2. Science (2021). paper; opened 2026-11-17 all records citing this
  10. BioPlex: download and release pages. Gygi Lab, Harvard Medical School. database, not peer reviewed; opened 2026-11-17 all records citing this
  11. Northeastern University Mass Spectrometry (NUMS) core: rates. Northeastern University (2025). organisation-page, not peer reviewed; opened 2026-11-17 all records citing this
Where it sits on the map
History of this record (3)

This record has been changed 3 times, on the dates below. The note kept against each change is written for the people who maintain the atlas, in their own shorthand, so it is not shown here; it is in the record itself.

Throughput class
medium
Emerging
no
Status
partly written
Last updated
2026-11-17
Record
content/assays/affinity-purification-mass-spectrometry.json