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.

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