pLLDT inconsistency between monomer and multimer

Hi, I encountered a strange inconsistency.

In a situation where a short AA sequence (e.g. <50AA) is folded as a monomer I often get high pLLDT values (i.e. >80 across >50% of the sequence, despite low MSA coverage). However, when the same sequence is co-folded with another chain (which usually has much higher MSA coverage), the corresponding pLDDT of the small chain of interest drops to unreliable/disordered values of 50-60 (in many cases much lower).

Can someone clarify this relative nature of pLDDT and how it can be interpreted?

Attached are 2 files of a single example, generated using ColabFold v1.5.2 (MMSEQ2-based), alone and with additional chain. Default parameters were used. I observed the same behavior in a non-MMSEQ2 version as well, so it's unlikely a pure MSA artifact.

Thanks!

[colabfold_mmseq2_multimer.zip](https://github.com/deepmind/alphafold/files/11585347/colabfold_mmseq2_multimer.zip)
[colabfold_mmseq2_monomer.zip](https://github.com/deepmind/alphafold/files/11585348/colabfold_mmseq2_monomer.zip)