Learn/Predict a structure
Fold a complex
A complex asks two questions: how does each chain fold, and where do the chains sit against each other? The second has its own confidence. For complexes the alignment is built per chain and then paired by species, so that covariation across the interface can be seen too.
Barnase and barstar
| Metric | AlphaFold 2 |
|---|---|
| pLDDT | 96.91 |
| pTM | 0.93 |
| ipTM | 0.92 |
The PAE's two blue squares on the diagonal are the two chains, each confident in itself. The off-diagonal blocks are the interface: blue means the model is sure where one chain sits relative to the other, while gold and coral mean uncertainty. ipTM summarises those blocks in one number; above 0.8 the arrangement is reliable, below 0.5 it is not.
Give the model the right number of copies
| Metric | Boltz-2 | Boltz-2 |
|---|---|---|
| pLDDT | 90.81 | 92.94 |
The transferrin receptor is a dimer on the cell. Folded as one copy it is a fine monomer with no interface to score; folded as two copies, the interface appears in the PAE and ipTM has something to say. The model folds whatever copies you give it, so the copy count is a decision, not a prediction.
Read more: AlphaFold-Multimer evaluation · Paired alignments in ColabFold