pLDDT (Predicted Local Distance Difference Test)

pLDDT is a per-residue confidence score ranging from 0 to 100. It predicts the local distance difference test (lDDT) score that the predicted structure would achieve against an experimental reference. A score of 100 indicates perfect local agreement, while lower scores signal regions of uncertainty. The metric is stored in the B-factor column of the output PDB file for direct visualization in molecular graphics software.

AlphaFold2's confidence bands guide practical usage:

• Very high (pLDDT > 90): High accuracy, suitable for detailed structural analysis and drug docking.
• Confident (90 > pLDDT > 70): Generally good backbone prediction, useful for fold classification.
• Low (70 > pLDDT > 50): Low confidence, treat with caution; may indicate flexible loops.
• Very low (pLDDT < 50): Strong predictor of intrinsic disorder; these regions likely lack a fixed structure in isolation.

pLDDT is explicitly a local superposition-free metric. It assesses the correctness of inter-atomic distances within a 15 Å radius around each residue, making it insensitive to rigid-body domain movements. A model can have high pLDDT scores in individual domains but a poor global arrangement. For domain-level orientation confidence, consult the complementary Predicted Aligned Error (PAE) matrix.

The pLDDT predictor is a separate output head of the AlphaFold2 structure module, trained self-supervisedly. During training, the predicted structure is compared to the ground truth, and the model learns to estimate the lDDT it would achieve. This makes pLDDT a well-calibrated prediction of error, not just an arbitrary internal score. It has been validated against experimental depositions in the PDB.

pLDDT serves a critical operational role during AlphaFold2's recycling phase. The per-residue confidence scores from one iteration are fed as input features into the next. This allows the model to iteratively focus its attention on low-confidence regions, progressively refining the structure over multiple recycling passes until the confidence estimates converge.

The standard visualization convention uses a color spectrum from blue to red:

• Blue: Very high confidence (pLDDT > 90)
• Cyan: Confident (90 > pLDDT > 70)
• Yellow: Low confidence (70 > pLDDT > 50)
• Orange/Red: Very low confidence (pLDDT < 50) This coloring is the default in the AlphaFold Protein Structure Database and is essential for rapid visual assessment of prediction quality.

A pairwise confidence metric that estimates the expected positional error between any two residues in a predicted structure. While pLDDT assesses local per-residue accuracy, PAE evaluates domain packing quality and relative orientation. A low PAE between two domains indicates high confidence in their relative positioning; high PAE suggests uncertainty in domain arrangement even if individual domain pLDDT scores are high. PAE plots are essential for interpreting quaternary structure predictions.

A scale-invariant metric for assessing topological similarity between two protein structures, designed to be independent of protein size. Unlike RMSD, TM-score provides an intuitive measure of structural fold agreement:

• >0.5: Indicates the same SCOP/CATH fold
• <0.2: Random structural similarity
• 1.0: Perfect match TM-score is heavily weighted toward shorter interatomic distances, making it more sensitive to global topology than local deviations.

The standard quantitative measure of the average distance between backbone atoms of superimposed protein structures. Calculated as the square root of the mean squared distances between equivalent atoms after optimal superposition. Key considerations:

• Units: Ångströms (Å)
• Typical values: <2 Å for high-accuracy predictions
• Limitation: Sensitive to domain movements and outliers; a single flexible loop can inflate global RMSD RMSD is most meaningful when comparing structures within the same fold and after careful structural alignment.

A widely used structure-validation software that analyzes all-atom contacts, hydrogen bonding, and backbone geometry to assess the physical realism of protein models. Key outputs include:

• Clashscore: Number of steric overlaps per 1000 atoms
• Ramachandran outliers: Residues in disallowed phi/psi regions
• Rotamer outliers: Unusual side-chain conformations MolProbity complements pLDDT by evaluating whether a high-confidence prediction is also physically plausible, not just statistically likely.

A continuous quality score for evaluating protein-protein docking predictions that combines three metrics into a single value ranging from 0 to 1:

• Fnat: Fraction of native interfacial contacts recovered
• iRMSD: Interface RMSD after optimal superposition
• LRMSD: Ligand RMSD after receptor superposition DockQ scores >0.8 indicate high-quality docking models. This metric is critical when pLDDT is used to assess multi-chain complex predictions from AlphaFold-Multimer.

A collection of structurally distinct states representing the intrinsic dynamic flexibility of a protein. A single high-pLDDT static prediction may not capture biologically relevant motions. Key concepts:

• Holostates vs. apostates: Bound and unbound conformations
• Cryptic pockets: Transient binding sites invisible in static structures
• Ensemble refinement: Using NMR, cryo-EM, or diffusion models to sample multiple states pLDDT reflects confidence in one conformation; ensemble approaches address the functional reality of protein dynamics.