Merge branch 'main' of ssh://github.com/aqlaboratory/openfold into main

README.md

@@ -89,9 +89,11 @@ where `data` is the same directory as in the previous step. If `jackhmmer`, `hhb
 ### Training
 
 After activating the OpenFold environment with `source scripts/activate_conda_env.sh`, install OpenFold by running
+
 ```bash
 python setup.py install
 ```
+
 To train the model, you will first need to precompute protein alignments. Create `mmcif_dir/` and download `.cif` files from the PDB (e.g., `wget https://files.rcsb.org/download/4DSN.cif`). Then run:
 
 ```bash

openfold/data/data_modules.py

@@ -449,7 +449,7 @@ class OpenFoldDataModule(pl.LightningDataModule):
 class DummyDataset(torch.utils.data.Dataset):
     def __init__(self, batch_path):
         with open(batch_path, "rb") as f:
-            batch = pickle.load(f)
+            self.batch = pickle.load(f)
 
     def __getitem__(self, idx):
         return copy.deepcopy(self.batch)

openfold/utils/loss.py

@@ -120,9 +120,6 @@ def compute_fape(
     return normed_error
 
 
-# DISCREPANCY: From the way this function is written, it's possible that
-# DeepMind clamped 90% of individual residue losses, not 90% of all batches.
-# We defer to the text, which seems to imply the latter.
 def backbone_loss(
     backbone_affine_tensor: torch.Tensor,
     backbone_affine_mask: torch.Tensor,
@@ -164,7 +161,6 @@ def backbone_loss(
             1 - use_clamped_fape
         )
 
-    # Take the mean over the layer dimension
     fape_loss = torch.mean(fape_loss)
     return fape_loss