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

Summary

Confidence metrics for structure prediction, such as pLDDT and pTM, correlate with prediction accuracy, even in the absence of coevolutionary data (1). This was also observed when modeling CDRH3 loops, which lack any meaningful coevolutionary signal (2). This has been suggested to indicate that an energy function is being learned.

Figures

Ref (1)

Ref (2)

See also

• Sequence- and structure-derived ML quality metrics from ML do not correlate with each other
• Protein folding neural networks make local optimizations in the absence of coevolutionary information

1.

Roney JP, Ovchinnikov S. State-of-the-Art Estimation of Protein Model Accuracy Using AlphaFold. Physical Review Letters. 2022;129(23). Available from: https://doi.org/10.1103/physrevlett.129.238101

2.

Chen H, Fan X, Zhu S, Pei Y, Zhang X, Zhang X, et al. Accurate prediction of CDR-H3 loop structures of antibodies with deep learning. eLife. 2024;12. Available from: https://doi.org/10.7554/elife.91512.4