AlphaFold PAE: Check Domain Placement Before Animating

Use AlphaFold PAE to check whether a predicted protein supports the relative placement of the regions you intend to show. First inspect local confidence with pLDDT, then inspect the PAE blocks connecting those regions. A convincing fold in each domain does not establish that the domains face each other correctly. If their placement is uncertain, keep that limitation visible in the figure or explain the domains separately.

This guide works through human calmodulin-1, entry AF-P0DP23-F1, using version 6 files downloaded on October 13, 2026. The file calculations below were executed locally. The website interactions follow current official documentation; they are not a record of a separately tested browser tutorial.

You need a browser, the correct protein accession and a place to keep the coordinates, confidence file and review notes together. The four CGI images here are conceptual illustrations. They do not depict the downloaded calmodulin coordinates or encode its measured confidence values.

A residue is one amino-acid unit in a protein chain. A domain is a compact structural region that can often be discussed as a folding unit. Predicted local distance difference test, or pLDDT, scores describe local model confidence on a scale from 0 to 100. Higher values indicate greater confidence.

The EMBL-EBI pLDDT guide uses above 90 for very high confidence, 70 to 90 for confident predictions, 50 to 70 for low confidence and below 50 for very low confidence. These categories help triage regions; they are not a percentage probability that a proposed mechanism is correct. Inspect the actual residues used in your argument rather than relying on the whole-protein average.

Predicted aligned error, or PAE, addresses a different relationship: expected positional error for one residue when the structure is aligned on another. Its unit is the ångström, written Å. Lower values mean greater confidence in that relative placement. A high-scoring local fold can therefore coexist with uncertain domain orientation.

The original AlphaFold study established prediction and confidence assessment against structural benchmarks. It did not turn every individual model into an experimentally observed conformation.

Open the P0DP23 entry and confirm calmodulin-1, Homo sapiens and the sequence you intend to illustrate. UniProt P0DP23 identifies the reviewed human sequence as 149 amino acids. This prevents a similarly named protein, fragment or different organism from entering the project unnoticed.

Record the entry identifier and model version before downloading. Our retrieved entry identifies the AlphaFold Monomer v2.0 pipeline and file version 6. The version number in a filename is a data revision, not a claim that the model was generated by a method called AlphaFold 6.

Save the coordinate file and PAE JSON from the same entry and version. The download documentation lists both formats and explains that entry pages display the latest version. Keep the original filenames in your project record even if you rename working copies for convenience.

In the entry’s Summary and Model Confidence tab, locate the sequence, 3D view and interactive PAE plot. The current interface guide documents these together. Start by identifying the sequence regions you plan to compare; do not choose a region merely because it looks attractive in the rendered model.

Each position in the PAE plot relates two residues. Follow the displayed axes and legend. Inspect the square block for each region against itself, then both rectangular blocks connecting the regions. The thin main diagonal compares residues with themselves and is not evidence that a whole protein is correctly arranged.

The PAE teaching guide explains how selecting a plot region highlights corresponding parts of the structure. Use that connection to check what a block actually represents. Low-error blocks within regions combined with larger errors between them support different confidence judgments for local structure and overall placement.

Keep both cross-region directions in your notes. PAE need not be symmetric, so replacing two rectangles with one average can hide information. A global summary also conceals whether the residues forming the proposed interface are the uncertain ones.

For a small, reproducible exercise, define window A as residues 6 to 72 inclusive and window B as residues 83 to 146 inclusive. These are analyst-selected comparison windows, not a new assignment of biological domain boundaries. The input contains 149 Cα atoms, one per residue, and a 149 by 149 PAE matrix.

Read one pLDDT value per residue from the Cα record’s B-factor field. This storage convention is documented in the AlphaFold output reference. It applies to these predicted coordinates; an experimental crystallographic B-factor has a different meaning. Averaging every atom would give residues with more atoms extra weight.

Calculate the arithmetic mean across the 67 values in A and the 64 values in B. For each PAE block, collect all row-column combinations in the specified ranges and calculate their median. Exclude self-pairs where row and column are the same residue. Preserve the stored row-column order and do not symmetrize the matrix.

For zero-based array indices, A occupies 5 through 71 and B occupies 82 through 145. The downloaded integer-valued PAE entries yield the following results. Their precision comes from the file format; it does not establish experimental error bars.

For these windows, the cross-region median errors are substantially larger than the within-region medians. This is a useful warning against presenting their relative placement as settled. The two means also conceal residue variation: A spans pLDDT 73.62 to 94.00 and B spans 74.69 to 95.88 in this file.

A defensible figure can show the two regions while explaining that their arrangement is a prediction with limited confidence. An alternative is to show separate close-ups when the argument concerns only their local folds. Neither option requires deleting the original whole-chain model.

The values do not tell you to move a domain by 23 Å, swing it through a particular angle or animate a repeated opening cycle. Those would introduce a physical trajectory that this confidence analysis never measured. A suggested transition needs independent structural or dynamical evidence before it can be described as biological motion.

Write one sentence for each shot before changing camera positions. This plan teaches how to inspect a prediction, so its only required motion is a camera move. Keep the molecular coordinates fixed unless another source justifies a conformational change.

Our surface-versus-ribbon guide helps choose which structural feature remains legible. A surface can make a region easy to recognize; a ribbon can reveal its internal fold. Neither representation increases confidence in the underlying coordinates.

Preserve a confidence-colored reference view even if the final presentation uses a different palette. If teal identifies a region in your animation, do not let a nearby legend imply that teal means high pLDDT. Use a caption or separate reference frame to explain the change in encoding.

A low-confidence extension can be hidden, muted or represented schematically when it distracts from the question. Record which residues were affected and keep the original available. Making a tail transparent changes what the audience sees; it does not validate the tail or demonstrate that it disappears biologically.

For a research figure, compare the sequence, experimental context and relevant structure records before drawing an interface claim. The biological-assembly guide addresses a separate check: whether the displayed collection of chains matches the intended complex.

Investigate a mismatch before replacing the model or writing the caption. First compare file versions and residue counts. If you crop coordinates to a domain but retain the full-length PAE matrix, keep an explicit mapping back to original sequence numbering.

Do not manufacture a universal acceptable-PAE threshold to finish the figure. The relevant confidence depends on the relationship being claimed. A domain overview and an atomic interface proposal require different scrutiny.

Use a short caption that identifies the source and the limit. For this exercise: “Human calmodulin-1 prediction AF-P0DP23-F1, version 6, retrieved October 13, 2026. Windows 6–72 and 83–146 were compared for illustration. Cross-window PAE is larger than within-window PAE; the view does not establish a fixed relative arrangement or a motion pathway.”

Add the representation, any hidden residues and the meaning of colors to your own caption. If a camera rotates around a fixed structure, say so. If a separate schematic explains the idea, identify it as a schematic rather than attaching the model accession in a way that suggests coordinate fidelity.

Keep the calculation recipe beside the source files. Another researcher should be able to recover the same windows, pair exclusions and rounded results without guessing which slide was exported first.