AlphaFold pLDDT and PAE Domain Confidence Inspector
Read the predicted structure confidence metrics the workspace shows for human EGFR (UniProt P00533): per-residue pLDDT (0–100) and the pairwise Predicted Aligned Error (PAE in Å). A single full-chain 3D ribbon can combine high-confidence domains with unanchored flexible linkers; separating local backbone accuracy (pLDDT) from inter-domain packing confidence (PAE) prevents misinterpretation.
Opens the predicted (AlphaFold DB) structure view for the full-length 1,210-residue human EGFR precursor, where per-residue pLDDT and the PAE matrix are displayed. The multi-domain receptor shows high-confidence domains joined by low-confidence linkers and a long disordered C-terminal tail.
Distinguish local folding confidence (pLDDT) from global domain packing (PAE)
When an experimental X-ray or cryo-EM structure is determined—such as PDB 3IKA covering the human EGFR tyrosine kinase domain—crystallographers typically express a truncated, stable domain construct and leave unresolved flexible loops absent from the coordinate file. A full-length sequence prediction for UniProt P00533 instead outputs Cartesian coordinates for every amino acid from residue 1 to 1,210. Treating all 1,210 residues as equally well-determined geometry leads to severe analytical errors; the kinase domain spans roughly residues 712–979, outside which confidence falls.
To read a predicted model rigorously, inspect the two orthogonal confidence metrics supplied with the prediction. First, per-residue pLDDT (stored in the B-factor column of AlphaFold PDB/mmCIF files on a 0–100 scale) estimates local Cα distance conservation: scores above 90 indicate high confidence in both backbone and side-chain rotamers, 70–90 indicates reliable backbone secondary structure, 50–70 warrants caution, and values below 50 typically flag intrinsic disorder rather than a physical ribbon path. Second, the 2D Predicted Aligned Error (PAE) matrix reports whether two high-pLDDT domains pack against each other in a defined orientation (low off-diagonal PAE in Å) or float independently across a flexible tether (high off-diagonal PAE).
Compare predicted apo chains against experimental ligand-bound PDB structures
Even within a globular domain where pLDDT exceeds 90, a sequence-based prediction represents a single unliganded conformational state conditioned on evolutionary multiple-sequence alignments. Protein kinases such as EGFR switch between active and inactive conformations via large movements of the activation loop (DFG motif) and αC-helix, and inhibitors selectively stabilize specific states.
In the protein workspace, switch to an experimental PDB structure where available to inspect resolved ligand contacts and chain coverage, and use the UniProt coordinate track to see which flanking domains or disordered tails lie outside any single construct.
Equations and variables
pLDDT confidence band classification
Band(i) = { ≥90: very high; 70–90: confident backbone; 50–70: low; <50: disorder/unstructured }
- pLDDT(i)
- Predicted Local Distance Difference Test Cα score at residue i on a 0–100 scale
- ≥90
- High local accuracy suitable for inspecting binding-site side-chain rotamers
- <50
- Ribbon coordinates should not be interpreted as a fixed 3D conformation
Asymmetric Predicted Aligned Error between residue x and alignment reference y
PAE(x, y) = E[ |rx,true − rx,pred| | aligned on local frame of residue y ]
- PAE(x, y)
- Expected positional error at residue x when structures are superposed on residue y (Å)
- rx,true, rx,pred
- True and predicted 3D coordinates in the reference frame of residue y (Å)
Scope and limitations
- High per-residue pLDDT inside two separate domains does not imply that their relative 3D orientation is known; inter-domain packing requires low off-diagonal PAE blocks.
- Regions with pLDDT < 50 often correspond to intrinsically disordered segments in isolation and frequently appear as long ribbon spaghetti loops that have no physical meaning.
- AlphaFold models predict a single static conformation without explicit ligands, cofactors, membranes, or post-translational modifications and are not validated to predict destabilizing point-mutation ΔΔG.
Sources and provenance
- UniProtKB human EGFR P00533CC BY 4.0
The interactive workspace displays the release of the retrieved record.
- Jumper et al. 2021 highly accurate protein structure prediction with AlphaFoldCC BY 4.0; pLDDT and PAE methodology definitions
- Varadi et al. AlphaFold Protein Structure DatabaseCC BY 4.0; database confidence conventions
Technical questions
What is the difference between pLDDT and PAE?
pLDDT (predicted Local Distance Difference Test, 0–100) measures how well a residue’s local neighborhood (within 15 Å) is predicted, regardless of where the rest of the protein sits. PAE (Predicted Aligned Error, in Ångströms) is a 2D matrix measuring the expected positional uncertainty of residue X when the model is aligned on residue Y, which is essential for judging whether two domains have a fixed relative orientation.
Why do full-length receptor models like EGFR show long unstructured ribbons around the kinase domain?
Full-length EGFR (1,210 residues) contains a signal peptide, extracellular ligand-binding domains, a single transmembrane helix, a cytoplasmic kinase domain, and a long C-terminal regulatory tail. In the absence of a lipid bilayer and dimerization partner, the C-terminal tail (pLDDT < 50) and inter-domain linkers lack a unique rigid fold, and the relative orientation between extracellular and intracellular domains has high PAE.
Can I dock small molecules into a pocket with pLDDT between 50 and 70?
No. Only regions with pLDDT > 90 have sufficiently reliable side-chain chi-angle predictions for detailed pocket geometry, and even at pLDDT > 90 an apo prediction may adopt an inactive DFG-out or closed loop state that differs from a holo ligand-bound crystal structure.
Connected workflows
- EGFR sequence and structure inspection
Inspect the EGFR P00533 UniProt sequence and its PDB structure mapping.
- KRAS G12D trial-record exploration
Explore clinical trial records for another oncogenic signaling target.
- 4PL IC50 curve fitting
Analyze biochemical dose-response curves for kinase inhibitors.