Summary
In protein structure prediction, uncertainty metrics can be repurposed as energy-like functions for ranking or optimizing candidate structures. AlphaFold without coevolutionary input ranks structural decoys with state-of-the-art accuracy (1), and diffusion-model scores can be interpreted as statistical potentials for structure ranking, mutation-effect prediction, and conformational sampling (2). The analogy concerns relative ranking and sampling objectives, not calibrated thermodynamic free energy; raw confidence scores still need not predict stability or binding affinity.
Related notes
- Diffusion-based protein structure prediction methods double as energy methods comparable to traditional force fields
- Protein structure prediction and design metrics don’t correlate with expression probability
- Confidence metrics for diffusion-based structure prediction methods can be improved with minimal changes to conditioning representations
- Including structure prediction confidence while training inverse folding improves sequence diversity but not sequence recovery
- Most ML quality metrics cannot effectively predict enzyme activity after controlling for similarity to native
- pLDDT correlates with number of homologous sequences provided during runtime
- Protein structure prediction and design confidence metrics do not correlate with binding affinity
- pLDDT and PAE inversely correlated with protein dynamics in dynamic naturally occurring proteins, but not de novo proteins
- pLDDT is inversely correlated with CDRH3 length
- Protein folding neural networks cannot predict protein stability
- Self-consistency perplexity is correlated with pLDDT
- AlphaFold3 ipTM can distinguish between antibody binders and nonbinders
- Inverse folding sequence perplexities correlate with Rosetta energies, forward folding TM-scores, and sequence recovery
- PAE weakly correlates with Ab-Ag binding
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.
Roney JP, Ou C, Ovchinnikov S. Protein Diffusion Models as Statistical Potentials. openRxiv; 2025. Available from: https://doi.org/10.64898/2025.12.09.693073