Summary
RMSD is a poor training objective for protein structure prediction. For proteins, FAPE is better (1). For ligands, the fraction of models predicted under a certain RMSD is better (2). However, it was used by (3) to train IgFold and was apparently effective.
1.
Baek M, DiMaio F, Anishchenko I, Dauparas J, Ovchinnikov S, Lee GR, et al. Accurate prediction of protein structures and interactions using a three-track neural network. Science. 2021;373(6557):871–6. Available from: https://doi.org/10.1126/science.abj8754
2.
Corso G, Stark H, Jing B, Barzilay R, Jaakkola TS. DiffDock: Diffusion Steps, Twists, and Turns for Molecular Docking. In: ICLR 2023. 2023. Available from: https://openreview.net/forum?id=kKF8_K-mBbS
3.
Ruffolo JA, Chu L-S, Mahajan SP, Gray JJ. Fast, accurate antibody structure prediction from deep learning on massive set of natural antibodies. Nature Communications. 2023;14(1). Available from: https://doi.org/10.1038/s41467-023-38063-x