devalab / DeepPocket

Ligand Binding Site detection using Deep Learning
MIT License
89 stars 26 forks source link

Why does DeepPocket not predict residues for some fpocket pockets? #35

Open JavierSanchez-Utges opened 4 months ago

JavierSanchez-Utges commented 4 months ago

Hi, I was trying DeepPocket on chain B of 1a5h and tried to segment ALL pockets predicted by fpocket. I have noticed that out of the 14 predicted by fpocket, only 3 have been segmented (I have removed the lines that break the loop when count >= 3. In this example, it is pockets 1, 3, and 10, which correspond to fpocket 2, 1, and 5, respectively. I have checked and there are no predicted residues:

Pocket 11 of 1a5h_B has [] predicted residues
Pocket 12 of 1a5h_B has [] predicted residues
Pocket 13 of 1a5h_B has [] predicted residues
Pocket 14 of 1a5h_B has [] predicted residues
Pocket 15 of 1a5h_B has [] predicted residues
Pocket 16 of 1a5h_B has [] predicted residues
Pocket 17 of 1a5h_B has [] predicted residues

And this is how masks_pred look like:

        [[0., 0., 0.,  ..., 0., 0., 0.],
         [0., 0., 0.,  ..., 0., 0., 0.],
         [0., 0., 0.,  ..., 0., 0., 0.],
         ...,
         [0., 0., 0.,  ..., 0., 0., 0.],
         [0., 0., 0.,  ..., 0., 0., 0.],
         [0., 0., 0.,  ..., 0., 0., 0.]],

        [[0., 0., 0.,  ..., 0., 0., 0.],
         [0., 0., 0.,  ..., 0., 0., 0.],
         [0., 0., 0.,  ..., 0., 0., 0.],
         ...,
         [0., 0., 0.,  ..., 0., 0., 0.],
         [0., 0., 0.,  ..., 0., 0., 0.],
         [0., 0., 0.,  ..., 0., 0., 0.]],

        [[0., 0., 0.,  ..., 0., 0., 0.],
         [0., 0., 0.,  ..., 0., 0., 0.],
         [0., 0., 0.,  ..., 0., 0., 0.],
         ...,
         [0., 0., 0.,  ..., 0., 0., 0.],
         [0., 0., 0.,  ..., 0., 0., 0.],
         [0., 0., 0.,  ..., 0., 0., 0.]]])

Any idea why this is? I would expect a pocket could be segmented out of each fpocket prediction. Thanks!