calcProteinInteractions() throws NumPy UFuncTypeError in ProDy 2.5.0

[HaleHasdemir]

Hi, I'm trying to follow the InSty tutorial with my pdb. But I have the following issue:

Calling calcProteinInteractions() causes a UFuncTypeError during salt bridge calculation:

File ~/ProDy/prody/atomic/select.py:1901, in Select._binop(self, sel, loc, tokens)
   1898 else:
   1899     # ndim must not be zero for in place operation
   1900     if ndim:
-> 1901         binop(left, right, left)
   1902     else:
   1903         left = binop(left, right)

UFuncTypeError: Cannot cast ufunc 'subtract' output from dtype('float64') to dtype('int64') with casting rule 'same_kind'

Environment:
ProDy version: 2.5.0 (installed from source)
NumPy version: 1.26.2
Python version: 3.11

[HaleHasdemir]

Update: I tried running the same calcProteinInteractions() function using the tutorial files and also with 1ubq.pdb (after adding hydrogens using PDBFixer), and it worked without errors. However, my own PDB file contains multiple chains, and that’s where the error occurs.

Could the presence of multiple chains be the reason for this issue?

[karolamik13]

It should work with multiple chains. As you can see here: http://www.bahargroup.org/prody/tutorials/insty_tutorial/multiplechainsPDB.html#extract-chain-chain-interactions-in-ensemble-or-trajectory-analysis

Can you provide an example PDB file so I could test it?

[jamesmkrieger]

What is your pyparsing version? If it’s too recent that may be why the selection is failing

[jamesmkrieger]

There’s a solution at https://techoverflow.net/2019/05/22/how-to-fix-numpy-typeerror-cannot-cast-ufunc-subtract-output-from-dtypefloat64-to-dtypeint64-with-casting-rule-same_kind/

Probably we should change the type of left or right or both to make it work

Can you show a longer trace back with what triggers this function?

[HaleHasdemir]

Hi all, thank you for the quick response!

I'm encountering the error while using this PDB file: https://www.rcsb.org/structure/6A78. I use the bioassembly 1. My current pyparsing version is 3.2.3. I’ve tried downgrading it incrementally down to 2.3.0, but the error persists.

[hyunp2]

Hello all, @HaleHasdemir and I resolved this issue.
We report the problem and solution here, partially inspired by @jamesmkrieger .

1. prody/atomic/selec.py line 1899 to 1903 has

else:
                ndim must not be zero for in place operation
                if ndim:
                    binop(left, right, left)
                else:
                    left = binop(left, right)

which leads to cast type issues for np.subtract

We just commented out the problematic parts to

else:
                # ndim must not be zero for in place operation
                # if ndim:
                #     binop(left, right, left)
                # else:
                left = binop(left, right) 

In other words, the inplace operate is a problem.

2. Also, we found another error which complains about indexing (i.e, line 895)
   indices = self.getIndices(atoms, selstr, **kwargs)
   It is because resid can be considered for integer 0 and positives, within ProDy.
   Our 6a78 has a chain H with a resid -1 and increasing.
   When I chose atomselection to protein and resid 0:10000 (i.e., 10000 being an arbitrary large number), we resolved this issue and can compute interactions.
   This is problematic for SaltBridge and Repulsive interactions.

Please advise and address this issues if our finding is of merit.
Thanks.

[jamesmkrieger]

Thanks

1. You can’t just comment out the part that doesn’t work in your particular case. There may be some reason that it’s there and some cases where it is needed. Please give us the full error or a pdb file so we can figure out what is triggering it. Still it’s good to see that this works as a guide.

2. There is a way of handling negative resid with back ticks that we can incorporate into insty. Thanks for reporting this