Global_score

[zjlclimb]

Does anyone know the mathematical meaning of global_score?

[matteoferla]

As far as I can tell, average loss values of the whole protein.
there are two scores, score and global_score in the output —the codebase has a native_score. The scores are a negative log likelihood for the loss function (one line in sup info).
They are calculated ( https://github.com/dauparas/ProteinMPNN/blob/main/protein_mpnn_utils.py#L39) using torch.nn.NLLLoss. That means low is good —so like the score in a golf game not a football, cricket etc game.
While high seq_recovery is good.

score --> arithmetic mean of the losses masked for design
global_score --> arithmetic mean of all losses (design+native)