seq_length dimension different to AlphaFold-multimer

[georgkempf]

In case of the original AlphaFold pipeline, the "seq_length" array in the feature_dict seems to get flattened in multimer mode (in monomer mode it stays a 1-D array). As far as I can see this does not happen in case of the OpenFold feature pipeline (where seq_length has the same dimension in monomer and multimer mode) and causes an incompatibility if one wants to use a features_dict from the original AlphaFold data pipeline. It might also be related to this issue: https://github.com/aqlaboratory/openfold/issues/111#issue-1252970018

Update: In case anyone else is trying to use the feature pipeline or features.pkl from original AF: There are some more differences in the feature_dict (int instead of float in case of "asym_id", "sym_id", "entity_id") which caused more errors downstream. I added now some code to my local installation to "convert" the dict from the original AF pipeline to make it compatible. Now the prediction worked.

[gahdritz]

Are you using the multimer branch? That's still experimental (we'll get to this eventually), and so small discrepancies like this are to be expected.

[moritzschaefer]

@gahdritz, really curious about the state of the multimer branch and whether there is a planned timeline for it to be finished.

Thank you for the great work!