[QUESTION] A question about antibody-antigen dataset in Figure 1.C

[ahmedselim2017]

Hi, first of all thank you for your work and for open sourcing it!

In the Figure 1.C of the whitepaper, the number of antibody-antigen complexes in the Abramson et al.'s dataset is given as 134. But in the Abramson et al.'s antibody-antigen dataset in the AlphaFold3's repository (https://github.com/google-deepmind/alphafold3/blob/main/docs/metadata_antibody_antigen.csv) includes 166 antibody-antigen interfaces in 71 unique biological assemblies.

Does the Figure 1.C use other antibody-antigen complexes other than the AlphaFold3's antibody-antigen dataset?

Also does the Figure 1.C predicts all of the antibody-antigen complex or just the 2 chain interfaces (heavy chain + antigen and light chain + antigen predicted separately)?

[jnwei]

@vinay-swamy @haewonc @gnikolenyi perhaps one of you could provide details on the antibody-antigen complex used in our evaluation in the white paper?

[vinay-swamy]

A couple of the larger complexes were too big for some of the other tools, so for a fair comparison we subset DeepMind's set to structures that were predictable with all tools. We predict the structure of the entire complex, then subset to the relevant interfaces defined in the the metadata you linked. @haewonc LMK if I missed anything.