Molecular Lipophilicity Potential

Elaine Meng
Aug-Sep 2016; 2019 update below

MLP (or MHP for molecular hydrophobicity potential) is a construct that spreads atomic values out in 3D, analogous to an electrostatic potential calculated from atomic partial charges. It is not clearly defined, so various different functional forms are used, as well as different sets of atomic parameters. Most of the emphasis in papers about the atomic contributions is on using them to predict the logP (octanol-water partition coefficient) of entire small organic molecules. Coloring/display of “hydrophobicity potential” is a later application.

Qualitative conclusion from the limited set of comparisons below: for protein depiction, a simple amino acid lookup may be be better than or at least as good as than a potential, given the latter's complexity, computational demands, and dependence on reasonable atomic values. See for example the result of simply averaging PLATINUM atomic values over residues without using a distance-dependent equation (below). The atomic values used by pyMLP and thus mlp in ChimeraX may not be that good; however, for the comparisons here I did not explore the different functional forms and other adjustable parameters in mlp (details...). The possible advantage of an atom-type-based potential is that it could be used on arbitrary organic compounds, but this has not been implemented in ChimeraX yet. Ideally we'd have both:

• a better, atom-type-based lookup table of values for MLP/MHP calculation
• a few different built-in amino acid hydrophobicity scales, say Kyte-Doolittle, Hessa, Moon-Fleming

Aug 17, 2016. Trying the PLATINUM server http://model.nmr.ru/platinum/ on membrane protein rhodopsin, 1hzx chain A:

input atomic values, range -1.76, 0.76 ("bfactor" in 1hzxAwithH_atomic.pdb)	input atomic values but residue avg, range -0.17, 0.11 (Chimera calc residue avg "bfactor")	output MHP at centers, range -1, 1 ("bfactor" in 1hzxAwithH_centers.pdb)	output MHP at surface, range -1.5, 1.83	same thing but coloring -1,0,1 (all others min,0,max)

... compare to current Chimera and ChimeraX options:

ChimeraX mlp default parms except color min,0,max min -45, max 25	ChimeraX mlp same except color -20,0,20 (default coloring range)	ChimeraX mlp default everything	Chimera kdHydrophobicity color min,0,max	Chimera wwHydrophobicity color min,0,max	Chimera wwHydrophobicity color -1,0,1	Chimera hhHydrophobicity color min,0,max *negative is more hydrophobic in this scale*	Chimera mfHydrophobicity (suggested by Oliver Clarke, see: http://plato.cgl.ucsf.edu/pipermail/chimera-users/2016-July/012562.html) color min,0,max *negative is more hydrophobic in this scale*

Files for/from the PLATINUM server (more details in their manual):

• input: 1hzxAwithH.pdb membrane protein rhodopsin (upload as "ligand")
• parameters: Ghose atomic values, function exp(-r/2), offset 0.03, low dot density
• output:
  • 1hzxAwithH_table.txt - text file listing atoms from the structure, their atom-type assignments, and atomic values for calculating MHP (before adding 0.03)
  • 1hzxAwithH_atomic.pdb - PDB file with *input* atomic values in bfactor column (after 0.03 was added)
  • 1hzxAwithH_centers.pdb - PDB file with *output* MHP values in bfactor column... I think. The manual says "surface MHP projected back onto atomic centers" but I don't know how the projection was done, and the values seem to have been either normalized or capped to the -1,1 range; image above sure looks similar to the residue-avg input values, but their histograms in Chimera Render by Attribute are different
  • 1hzxAwithH_surf.pdb - giant PDB file with surface dots as atoms, *output* MHP values in bfactor column

• New (Ghose, 1998) [paper online]
• Obsolete (Viswanadhan, 1989)

Distance function:

• exponential [exp(-r/2)]
• exponential [exp(-r)]
• hyperbolic
• Fermi-like

← ChimeraX Update (2019)

Due to some limitations of the original set of atomic lipophilicity values from pyMLP, most noticeably asymmetry and sign difference in the values for ASP/GLU carboxylate oxygens, I investigated using the atomic lipophilicity values in the Ghose paper instead. These were developed for ALOGP/CLOGP calculations and are used by the PLATINUM webserver.

This paper lists very many atom types, much more subdivided than the ChimeraX atom types (for example, nine different types for just H bonded to C!). It's not feasible for us to encode the chemical rules to recognize all of these types currently, and we want to allow MLP coloring without requiring hydrogen addition. Thus, I manually created a lookup table with the same amino acids as the original pyMLP set in which the values for the (inferred) attached hydrogens were added to the respective heavy-atom values. I'll call this set of atomic parameters ghose-united. To the original set of residues, I added a few more types that could occur within a protein chain: MLE and the peptide-capping residues NH2, NME, and ACE, as well as UNK (backbone only, sometimes used for lower-resolution structures in which the amino acid type cannot be determined from the density).

I also made a ghose-united-shifted set incorporating the 0.03 shift per atom recommended in the Platinum manual, accounting for the 0-3 hydrogens that had been united with each heavy atom. These files can be tested by being substituted for the file mlp.py (keeping that name) in the ChimeraX download.

Numerical results, images, and conclusions as to what we should use are given below.

Assumptions and judgment calls:

• Cys assumed to be SH, not disulfide-bonded (although the difference in the SG atomic value would be small)
• His, Arg, Lys sidechains assumed to be always positively charged
• Asp, Glu sidechains assumed to be always negatively charged
• there was no carboxylate oxygen type, so I averaged the values for carbonyl O and O^– types to give –0.4087 each (symmetric and both hydrophilic, in contrast to –0.68 and +0.53 in the original parameter set)
• peptide backbone atoms are assumed to be in-chain, i.e., there is no detection of the N-terminal N and changing it from amide to NH3+ type
• some atoms fit the description of more than one type, so I had to just pick one or the other (e.g. TRP CE2 is both R--CX--R and R--CR···X, where R is any group linked through carbon, X is any heteroatom, -- is an aromatic bond, and ··· is a pyrrole-like aromatic bond)

I tried the ChimeraX mlp command with these different atomic-value sets on several proteins, including:

• transmembrane proteins: 1hzx chain A (biological unit is monomer) 1a0s chain Q (biological unit is trimer), 1bxw, 6o2p
• soluble (globular) proteins with a hydrophobic pocket: 3mhc, 3w7f

structure	min, mean, max on surface (rounded)
	original	ghose-united	ghose-united-shifted
1hzxA	-45, -5, 25	-27, 0, 24	-21, 4, 30
1a0sQ	-48, -8, 25	-34, -5, 24	-28, -1, 28
3mhc	-45, -9, 23	-31, -5, 23	-26, -2, 28
3w7fA	-45, -7, 25	-26, -4, 25	-21, 0 30
1bxw	-50, -9, 22	-28, -4, 23	(not tried)
6o2pA	-48, -6, 26	-29, -2, 26	(not tried)
6o2pB (has only UNK res, bb atoms)	0,0,0	-24, -11, -2	(not tried)

1hzx chain A, default coloring range (-20,20) except where noted otherwise:

original	ghose-united	ghose-united (range –15,15)	ghose-united-shifted
ghose-united (same as above except Asp/Glu highlighted)

A few more of the protein examples, default coloring range (-20,20):

1bxw		6o2p		3mhc
original	ghose-united	original	ghose-united	original	ghose-united

Conclusions

Quite consistently on a variety of proteins, the original parameters give a minimum of negative 40-50, mean of negative 5-10, and max in the 20s, whereas the ghose-united parameters shift the minimum to negative 25-35 and the mean closer to zero, but give about the same maximum as the original parameters. Coloring over the same default range (-20,20) gives the same overall impression, with the same regions coming out as hydrophobic, and not surprisingly, a more reasonable result for Asp/Glu sidechains and UNK/backbone-only residues. I recommend using the ghose-united parameters as the new default and keeping the default coloring range the same. The additional ad hoc shift (ghose-united-shifted parameters) doesn't seem to add anything to the analysis. Anyone who prefers a more saturated coloring could use a more restricted coloring range (e.g. -15,15) but I think the -20,20 range provides a better sense of the range of values anyway. (A frequent problem in paper figures is the use of oversaturated coloring for electrostatic potential.)