[apbs-users] Strange Output for APBS Binding Energy Calculations

[David Rosenman]

Hello all,

First, an apology in advance if any of this has been discussed in other topics on the mailing list, I am having trouble reading many of the posts in the archive.

I'm trying to measure the electrostatic energy of binding for different homodimeric binding poses of a particular ~400 residue protein. To accomplish this, I have been using PQRs generated by PDB2PQR in tandem APBS input akin to the tutorial here:

http://www.poissonboltzmann.org/examples/Protein-Rna_Tutorial/

the major differences being, only the dimer and monomer energies are calculated, and the final line is "print elecEnergy dimer - monomer - monomer end", and, of course, the grid dimensions established by {dime, cglen, fglen}. Here, I am using the grid dimensions recommended by PDB2PQR's APBS input file, for both the dimer and monomer calculations. 

The results I produce have a smooth variation with ionic strength similar to the results in the tutorial, but the first red flag for me is that I am getting energies that vary from +/- 200-2000 kJ/mol. These results can be up to 2 orders of magnitude from the results we expect.

Next, I am interested at how this binding energy varies with distance between the two monomers at a given ionic strength (here, 0.15 M). So, I generated PDBs where I separate the COMs of the proteins in 5 Angstrom increments on a fixed axis, with no steric clashes, up to say, 50 Angstoms (plus the starting separation). Whether initially repulsive or attractive, I would expect the magnitude of the energy to fall off with distance.

In doing this calculation, I took two approaches to setting up the grids. The first was to have all calculations on the largest grid established by the PDB2PQR's APBS results for the most separated monomers, so the grid is the same for all distance trials. The second was to use the PDB2PQR grid recommended at each distance.

Unfortunately, the results for either approach produce what seem to be random fluctuations around zero for all points, with either positive or negative energies with 10^2/10^3 kJ/mol magnitudes. There are no discernible trends, and in fact, some of the largest energies come at the greatest monomer-monomer distances (that produce data, see P.S. below).

Also, if I do the calculation where I use the same grid for all distances, the energy at, say, the first point is drastically different then if I use the recommended grid.

What all of this suggests is that there is a dramatic dependence for the binding energy on the grid dimensions, and on the input file - way more than the subtle changes one might expect. I'm worried at this point I am just studying random noise.

Thank you very much in advance for your help. Let me know if I can provide any more information. I wanted to keep this mail simple, but I can produce any of the APBS input/output upon request. The only thing I can't provide are the PDB/PQR files, due to confidentiality reasons.

Best regards,

David J. Rosenman

Postdoctoral Researcher

Lenhoff Lab

University of Delaware

P.S.

This is less important to me, but APBS also fails for certain poses for the largest 2 trials (45 Angstrom/50 Angstrom). They are using the same grid as the closer distances, so I don't know why. No error message is produced, only the first few lines of APBS output is produced, and the PB calculations are entirely skipped.

[Baker, Nathan]

Hi David --

It is best to use the same grid for all calculations.  However, you need to ensure that this is grid does not just have the same length and spacing but also the same center (e.g., centered on one of the molecules).  

What is your grid spacing?  Ideally, it would be under 0.2 Å for this type of calculation.

I hope this helps.

Thanks,

-- 

Nathan Baker

Director, Advanced Computing, Mathematics, & Data Division — Pacific Northwest Natl Lab

Visiting Faculty, Division of Applied Mathematics — Brown University

+1-509-375-3997 — http://go.usa.gov/czB5G

[David Rosenman]

Hello Nathan,

Thank you for your response.

I've been using the same grids for monomer and dimer components of the BE calculation, centered on the same molecule (the dimer in both cases). For the distance calculations, however, I realize that the dimer "molecule" is changing in each run as distance increases, so that may be introducing error.

So, I redid those, this time centering on the constant monomer. However, this did not fix the issue; I still get BE's that randomly fluctuate around zero versus distance. (Though, I will also note that the exact values are substantially different from what we got when we centered on dimers).

Then I tried to consider your second point in your email about grid spacing. The calculations I've done up to now have been using the PDB2PQR recommended grid setups. In general, these seem to yield a grid spacing of ~0.7 angstroms on the coarse grid and ~0.5 angstroms on the fine grid.

To try to address that issue, I tried to multiply dime values by ~3 (while satisfying the c*l^(nlev + 1) + 1 requirement) to get a fine grid spacing of <0.2 angstroms. However, when I do this, the program aborts without doing any PB calculations and with no error message, similar to my P.S. in my first email. Any idea what could be going on here?

Thank you very much for your help!

Best regards,

David

[Star, Keith]

Hi David,

If you could send us example files (.in, .pqr, etc.), that cause APBS to silently fail, then we could investigate the problem, and release a fix. 

Thanks,

 - Keith

------------------------------------------------------------------------------
Mobile security can be enabling, not merely restricting. Employees who
bring their own devices (BYOD) to work are irked by the imposition of MDM
restrictions. Mobile Device Manager Plus allows you to control only the
apps on BYO-devices by containerizing them, leaving personal data untouched!
https://ad.doubleclick.net/ddm/clk/304595813;131938128;j

_______________________________________________
apbs-users mailing list
apbs-...@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/apbs-users

[David Rosenman]

Hello Keith,

Thanks for looking into this for me. I'll post an example of the APBS input that causes the failure below, but I unfortunately cannot provide the PDB/PQR files due to confidentiality reasons.

read

    mol pqr dimer-0.pqr  # Dimer 

    mol pqr monomer.pqr  # Single molecule

end

elec name dimer  

mg-auto  

dime 769 961 769

cglen 190.5581 242.5968 183.1801

fglen 132.0930 162.7040 127.7530  

cgcent -24.137 -31.571 -24.033

fgcent -24.137 -31.571 -24.033   

mol 1  

npbe  

bcfl sdh  

pdie 2.0000

sdie 78.5400

ion charge 1 conc 0.15 radius 1.02

ion charge -1 conc 0.15 radius 1.81   

srfm mol  

chgm spl2  

sdens 10.00  

srad 1.40  

swin 0.30  

temp 298.15  

calcenergy total  

calcforce no  

end

  

elec name monomer  

mg-auto  

dime 769 961 769

cglen 190.5581 242.5968 183.1801

fglen 132.0930 162.7040 127.7530

cgcent -24.137 -31.571 -24.033

fgcent -24.137 -31.571 -24.033   

mol 2  

npbe  

bcfl sdh  

pdie 2.0000  

sdie 78.5400 

ion charge 1 conc 0.15 radius 1.02

ion charge -1 conc 0.15 radius 1.81  

srfm mol  

chgm spl2  

sdens 10.00  

srad 1.40  

swin 0.30  

temp 298.15  

calcenergy total  

calcforce no  

end  

 

print elecEnergy dimer - monomer - monomer end  

quit

Some explanations:

1. dime was originally "257 321 257". I tried to ~triple it as explained in the previous email, and that causes the blank output.

2. cgcent/fgcent originally centered on "mol 1". As explained in the previous email, I realized this dimer's center is changing with increasing separation distance, so I provided a manual center instead (I later noticed that centering on mol 2, the monomer, does not work for larger distance separations since they fall out of the grid. Here, I supplied the center of the largest distance dimer that produced output as the manual center for all runs).

3. cglen/fglen come from PDB2PQR for the largest separation dimer (50 Angstroms plus the original separation).

Here's what the output looks like:

Running apbs on apbs-0.in

----------------------------------------------------------------------

    APBS -- Adaptive Poisson-Boltzmann Solver

    Version 1.4.2

    

    <APBS copyright information here, omitted for brevity>

This executable compiled on Jan 11 2016 at 18:05:36

Parsing input file apbs-0.in...

read

Parsed input file.

Got paths for 2 molecules

Reading PQR-format atom data from dimer-0.pqr.

  13336 atoms

  Centered at (-1.030e+01, -1.583e+01, -1.569e+01)

  Net charge 2.06e+00 e

Reading PQR-format atom data from monomer.pqr.

Then it aborts, right there.

I hope that is enough to start on; I can understand if this is impossible to debug without the original PQR files. 

If so, let me know, and I can try to run APBS on a different set of coordinates, and see if I get the same errors. Then, I can provide those coordinates for debugging.

Best regards,

David

[Baker, Nathan]

Hi David -

I think Keith was helping you with this.  Do you and he make any progress?

[Star, Keith]

Juan and I have both been looking at it. I'm also looking at a similar issue, possibly related, where a user is utilizing very large dime values. 

 - Keith