About distogram format about alphafold HOT 2 CLOSED

Hi,
Your interpretation of the output is almost correct, however these are raw logits rather than probabilities. If you would like to interpret these as probabilities you should apply a softmax in the final dimension, e.g. jax.nn.softmax(logits, axis=-1) . The intervals are half-open (open on the left) and the bin edges represent the interval boundaries (hence n-1 bin edges).
The code for mapping from raw distances to distogram bins can be found here:
https://github.com/deepmind/alphafold/blob/main/alphafold/model/modules.py#L1387-L1396

We will also try and update the README to make this clearer for future users, thanks!

from alphafold.

Excellent, thanks!

from alphafold.