1.preprocessing_sp.py error

[kimdn]

When

(iNNterfaceDesign) [kimd999@deception02 run_PepSep]$ cat PepSep.input
pdb: 3ztj_ABCDEF
# same as PepBB

anchor_res: 1570-1574
# same as Manual.docx

it resulted in

I expect that now it will be trivial for you to update
1.preprocessing_sp.py
as you did for
1.preprocessing.py

[rsyrlyb]

PepSep requires 3ztj_ABCDEFHL.pdb from test_examples/PepSep. 3ztj_ABCDEF.pdb does not have residues 1570-1574

[rsyrlyb]

try inputs from test_examples/PepSep

[kimdn]

Ok, I just followed

as in Manual.docx.

Now using files from test_examples/PepSep
PepSep seems to have ran well.

[rsyrlyb]

I agree that Manual if not perfect yet. I'll change it tomorrow.

[kimdn]

4.amn_sampling.py ran as well.

So I've run both PepBB and PepSEP thanks to you.

Just, I was confused what is different between PepBB and PepSEP.

What PepBB does

• construct_binding_sites (which extracts binder binding part from input pdb?)
• generate bb of binder
• idealize bb of binder
• sequence design (similar to FastDesign, but uses LSTM instead?)

What PepSEP does

• construct_binding_sites (which extracts binder binding part and binder itself from input pdb?)
• sequence design (similar to FastDesign, but uses LSTM instead?)

So to me, the only difference between the two is
PepBB does generate bb of binder on top of what PepSEP already performs, right?

[rsyrlyb]

What PepBB does

construct_binding_sites (which extracts binder binding part from input pdb?)
generate bb of binder - PepBB does using LSTM
idealize bb of binder
sequence design - PepSep does (similar to FastDesign, but uses LSTM instead?)

What PepSEP does

construct_binding_sites (which extracts binder binding part and binder itself from input pdb?)
sequence desig - PepSep does (similar to FastDesign, but uses LSTM instead?)

other operations are conducted using pyrosetta including construct_binding_sites (selecting residues around anchor residue) and idealize using Idealize mover.
In both cases aas are predicted using PepSep.
We did not use FastDesign for PepBB backbones because FastDesign works bad on NN-generated backbones with little imperfections here and there which can be removed later, after relaxation with mapped predicted aas. You can read about it in biorxiv paper. PepSep was trained for PepBB backbones basically. But PepSep can be applied on normal backbones too, which is shown in What PepSEP does