pLDDT

(LDDT redirects here) pLDDT (predicted local distance difference test) is a confidence metric used by neural networks for protein structure prediction. It captures the per-residue accuracy, both in terms of neighborhood and side chain rotamer. It was first directly integrated into structure prediction by AlphaFold2 at the per-residue level and has been widely adopted since. AlphaFold3 adopted per-atom pLDDT.

Figure from (1)

Notes

When clustering predicted protein structures, sparse clusters tend to have lower pLDDT (2). This was found to be independent of MSA depth.
pLDDT correlates poorly with GDT-TS among AlphaFold2 models in CASP15. This was observed in a repeat that used deeper MSAs (3).
De novo sequences designed by inversion with high pLDDT were found by ESM to have high perplexity (4).
While the default pLDDT is not continuously differentiable and thus unsuitable for training, (5) use a modified version that can be used as a loss function.
pLDDT can be used as spatial restraints in biomolecular simulations: $E_{p L DD T} = \sum (\frac{d _{i}}{1.5 * exp ( 4 * ( 0.7 - p L DD T _{i} ) )})^{2}$ The equation was originally presented by Hiranuma et al. (6) and was used by del Alamo et al. (7) as coordinate constraints in Rosetta when refining AlphaFold2 models.

Terwilliger TC, Afonine PV, Liebschner D, Croll TI, McCoy AJ, Oeffner RD, et al. Accelerating crystal structure determination with iterative AlphaFold prediction. Acta Crystallographica Section D Structural Biology. 2023;79(3):234–44. Available from: https://doi.org/10.1107/s205979832300102x

Nomburg J, Doherty EE, Price N, Bellieny-Rabelo D, Zhu YK, Doudna JA. Birth of protein folds and functions in the virome. Nature. 2024;633(8030):710–7. Available from: https://doi.org/10.1038/s41586-024-07809-y

Lee S, Kim G, Karin EL, Mirdita M, Park S, Chikhi R, et al. Petascale Homology Search for Structure Prediction. openRxiv; 2023. Available from: https://doi.org/10.1101/2023.07.10.548308

Verkuil R, Kabeli O, Du Y, Wicky BIM, Milles LF, Dauparas J, et al. Language models generalize beyond natural proteins. openRxiv; 2022. Available from: https://doi.org/10.1101/2022.12.21.521521

Trinquier J, Petti S, Park S, Herath K, van Kempen M, Feng S, et al. SoftAlign: End-to-end protein structures alignment. openRxiv; 2025. Available from: https://doi.org/10.1101/2025.05.09.653096

Hiranuma N, Park H, Baek M, Anishchenko I, Dauparas J, Baker D. Improved protein structure refinement guided by deep learning based accuracy estimation. Nature Communications. 2021;12(1). Available from: https://doi.org/10.1038/s41467-021-21511-x

del Alamo D, DeSousa L, Nair RM, Rahman S, Meiler J, Mchaourab HS. Integrated AlphaFold2 and DEER investigation of the conformational dynamics of a pH-dependent APC antiporter. Proceedings of the National Academy of Sciences. 2022;119(34). Available from: https://doi.org/10.1073/pnas.2206129119

Quartz 4

Explorer

pLDDT

Notes

Antibody-antigen modeling by diffusion-based structure prediction is data-limited

Protein structure prediction and design confidence metrics do not correlate with binding affinity

AlphaFold3 uses cross-distillation from AlphaFold2 to avoid hallucinating secondary structure in low-pLDDT regions

Fine-tuning AlphaFold for MHC prediction by adding Evoformer layers improves RMSD and pLDDT

Fine-tuning base models on test cases can improve the performance of variant effect and structure prediction

Including structure prediction confidence while training inverse folding improves sequence diversity but not sequence recovery

Low-pLDDT hallucinated models from AF3-generation protein structure predictors are designable

Most ML quality metrics cannot effectively predict enzyme activity after controlling for similarity to native

Natural sequences have higher pTM but lower pLDDT than de novo sequences

PLM-designed sequences match the distribution of fitness values, lengths, and structure prediction confidence of natural sequences

Protein design using sequence-based models does not benefit from scale

Protein folding neural networks cannot predict protein stability

Protein structure prediction and design metrics don't correlate with expression probability

Randomly masking residues prior to running ESMFold allows false positives to be identified more effectively

Reinforcement learning outperforms fine-tuning in property-guided inverse folding

Self-consistency perplexity is correlated with pLDDT

Structure prediction uncertainty metrics as energy functions

pLDDT and PAE in protein folding neural networks are correlated

pLDDT and PAE inversely correlated with protein dynamics in dynamic naturally occurring proteins, but not de novo proteins

pLDDT correlates with number of homologous sequences provided during runtime

pLDDT is inversely correlated with CDRH3 length

pTM is less lenient than pLDDT toward structural novelty when predicting alternate conformations

Flow matching and diffusion perform comparably for biomolecular structure prediction

Confidence metrics for diffusion-based structure prediction methods can be improved with minimal changes to conditioning representations

Confidence metrics from structure prediction correlate with accuracy even in the absence of coevolutionary data

DiffDock confidence is inversely correlated with RMSD

High-pLDDT designs can be insoluble

Inverse folding methods outperform sequence design by hallucination

Sequence- and structure-derived ML quality metrics from ML do not correlate with each other