How to choose seeds ？

[chunqing403]

When I predict a protein-peptide complex, I use different seeds to run 10 times, and I get different IPTMs and PTMs, some IPTMs are less than 0.8, and some IPTMs are greater than 0.8. Is this predicted complex credible or not?
Or when we use 10 seeds and get 50 predicted complexes, we only select the complex with the highest ranking_score, and further judge the credibility of the complex through indicators such as IPTM and PAE?

[joshabramson]

We suggest using chain_pair_iptm to rank specific chain-pair interactions - see the output documentation.

Correlations between chain_pair_iptm and accuracy are shown in Figure 4a of the AF3 paper. It doesn't break out scores for protein-peptide specifically, just general protein-protein, but 0.8 is considered a good iPTM score.