← Back to reading

Reading protein structures and assessing predictions

In brief

Before rotating a model, identify what it represents. An experimental structure is based on measurements; a computed model predicts atomic positions. For the former, check the experimental method and its relevant quality measures. For the latter, check the modelling method and confidence estimates. RCSB presents these categories separately. An attractive three-dimensional view cannot replace that information. [1]

Evidence for this statement

Source-based explanation

Where to check this statement

guide-source-1
Definitions for assessing experimental and computed structures, not independent validation of a particular model.
Read the section →
Save, investigate and check this version

Editorial draft · human review pending

Ask about this material
On this page

Read and explore

Next, connect the structure to the sequence. Check organism, chain, isoform and represented region. A structure may cover only part of a protein, with residue numbering different from its UniProt record. SIFTS supplies residue-level mappings between PDB and UniProt. A missing mapping should be explicit; selection must not jump to a similar residue number in a different chain.

[2]
Evidence for this statement

Source-based explanation

Where to check this statement

guide-source-2
Description of residue mapping, not revalidation of a particular alignment.

For AlphaFold models, pLDDT describes local confidence for individual residues on a 0–100 scale. A low score can reflect flexibility or insufficient information for prediction; it does not establish lack of function. High confidence in individual domains does not establish their relative placement.

[3]
Evidence for this statement

Source-based explanation

Where to check this statement

guide-source-3
Definitions of local and relative-placement confidence. No particular structure was independently validated.

PAE addresses relative placement: it estimates the positional error of one residue when aligned on another, in ångströms. Low PAE between selected parts supports confidence in their relative positioning. This answers a different question from pLDDT. If PAE is unavailable, the model's local-confidence colours alone do not establish the arrangement of two domains.

[4]
Evidence for this statement

Source-based explanation

Where to check this statement

guide-source-4
Definitions of local and relative-placement confidence. No particular structure was independently validated.

Try the route

Try it. Open an available structure from a protein record and record its identifier, type and source. Select a sequence region and check which chain and residues are highlighted. Compare structural coverage with the full protein length. For a prediction, find a region of higher and lower confidence. If the measure is unavailable, record that rather than judging confidence from shape.

Evidence for this statement

EcoGenesis learning activity

    What confidence establishes

    Check your understanding. High pLDDT alone does not establish partner binding, enzyme activity or a variant's effect. Describe exactly what you checked: identity, coverage and quality of the selected region. Save the model, chain, region and data versions.

    Evidence for this statement

    EcoGenesis learning activity

      Coverage and confidence

      Missing coordinates and low confidence are different states. Values and colours belong to the selected source model.

      Evidence for this statement

      EcoGenesis learning activity

        Teaching diagram of coverage and confidence — no particular protein data
        1. Are coordinates present?

          Hatched areas of the teaching strip represent missing model coordinates. This is separate from confidence in coordinates that are present. The strip describes no particular protein; segment width is not a length.

        2. Prediction confidence

          The two outlines indicate different local-confidence levels without invented pLDDT values. PAE addresses a different question: the relative placement of model parts. Neither score alone establishes function.

        3. Residue mapping

          For a real structure, map the chain and residue numbers to the sequence. SIFTS helps trace that correspondence; missing coordinates must not silently be filled with predicted geometry.

        Worked result: high local confidence can coexist with uncertain placement

        Teaching case: two domains have high local pLDDT, but the PAE between them is high. A suitable conclusion is that their individual folds are more confident than their placement relative to one another. It is unsuitable to call the whole interdomain arrangement experimentally verified. If the structure lacks coordinates for a region, record missing coverage separately from a region modelled with low confidence.

        Evidence for this statement

        EcoGenesis

          Continue with an organism

          The photographs and molecular record come from separate source materials.

          Norway spruce →Arabidopsis thaliana →

          Your research question

          Save the article to return to your question and this source version.

          Plan a test

          This is your research plan. Completing these fields does not confirm a hypothesis.

          Ask about this material

          Sources and versions

          Editorial draft · human review pending

          1. RCSB structure quality

            2026-09-19 · Terms of use on source site

            Provenance and original
            Source version
            Page checked 19 September 2026; no immutable provider release asserted
            Source reading scope
            Definitions for assessing experimental and computed structures, not independent validation of a particular model.
            Location in the original source
            Definitions for assessing experimental and computed structures, not independent validation of a particular model.
          2. PDBe SIFTS

            2026-09-19 · Terms of use on source site

            Provenance and original
            Source version
            Page checked 19 September 2026; no immutable provider release asserted
            Source reading scope
            Description of residue mapping, not revalidation of a particular alignment.
            Location in the original source
            Description of residue mapping, not revalidation of a particular alignment.
          3. Understanding pLDDT

            2026-09-19 · Terms of use on source site

            Provenance and original
            Source version
            Page checked 19 September 2026; no immutable provider release asserted
            Source reading scope
            Definitions of local and relative-placement confidence. No particular structure was independently validated.
            Location in the original source
            Definitions of local and relative-placement confidence. No particular structure was independently validated.
          4. Reading PAE

            2026-09-19 · Terms of use on source site

            Provenance and original
            Source version
            Page checked 19 September 2026; no immutable provider release asserted
            Source reading scope
            Definitions of local and relative-placement confidence. No particular structure was independently validated.
            Location in the original source
            Definitions of local and relative-placement confidence. No particular structure was independently validated.
          Inspect structured data
          Content version
          e2a3646f46fcea0100e6ab8a117be2b838c43ebf4deb86a10a7fff9422168d7c
          Source version
          guides-20260919@cd1e4a0a6f0c71995fb6e076b59bd2ee930c2104150357fab50fc56f5f6e80fe
          Updated
          2026-10-01