Multimer ipTM big variance among five models

[yifeng-evo]

Dear developers,

I am using Alphafold Multimer (2.3.1) to identify a potential receptor for my ligand protein. I used "-l 1" so it only predicts 5 models. I am pairing my ligand with a list of putative receptors for multimer prediction. Most of the time I saw ipTMs from all 5 models are similar, but I also saw cases like the 5 ipTMs can vary a lot: 0.3019,0.4814,0.8192,0.3361,0.8782. Should I trust the model with best ipTM, or that I should look at the average from 5 models? What is the meaning of the big variance?

Thank you!