It seems the chain order used for structure prediction affects the confidence metrics. Usually, the effect is subtle, and the difference in the predicted structures is negligible, but sometimes it is noticeable, as in the example below. This is an artificial peptide-RNA complex; both confidence and structure are different depending on chain order.
>peptide
YNFTDFMRQLLQ
>rna
CAUAAGCCUAGCGCUUCUGG
If predicted as rna/peptide
pTM = 0.53
Mean pLDDT = 0.85
if peptide/rna
pTM = 0.60
Mean pLDDT = 0.90
I used esmfold2-2026-05 with 10 loops for this example
It seems the chain order used for structure prediction affects the confidence metrics. Usually, the effect is subtle, and the difference in the predicted structures is negligible, but sometimes it is noticeable, as in the example below. This is an artificial peptide-RNA complex; both confidence and structure are different depending on chain order.
If predicted as rna/peptide
pTM = 0.53
Mean pLDDT = 0.85
if peptide/rna
pTM = 0.60
Mean pLDDT = 0.90
I used esmfold2-2026-05 with 10 loops for this example