3
$\begingroup$

Are there known dimensional measurements for the classification of folded proteins given a starting chunk/domain as defined by something like the clustering functionality of MSM Builder?

Examples of what I would be looking for would be dimensions such as:

  • pH

  • Salinity

  • Dissolved Sugars

  • Reducing Agent Concentration

  • Ambient Electronegativity

With corresponding scales and precisions such as:

  • logorithmic vs linear

  • step size

  • distribution (e.g. dissolved/flat vs fractal/inclusions)

The above parameters would be included in folding software after a known starting conformation provided by MSM Builder or similar chunking software to derive a GUID for proteins without (or with extremely minimal and tuneable) collisions?

For instance, if hundreds of GUIDs on average end up equating to the same folded protein that's fine, but if one GUID could map to two or more thermodynamically stable folded proteins at greater than some incredibly small (e.g. 0.0000001%) then it wouldn't fit what I'm looking for. I know this is inherently NP-hard, but the complete ID would be something like:

{
    sequence,
    chunkId,
    pH,
    salinity,
    dissolvedSugar,
    reducingAgentConcentration
}

So I would expect the chunkId and dimensional parameters to play a significant role in reducing collisions for different starting parameters.

It's worth noting I'm not seeking some magic bullet where you type an ID and get a fully folded protein, just a reliable system of GUIDs with enough starting information embedded in the ID to start the folding process.

$\endgroup$
3
$\begingroup$

CATH

The CATH database classifies protein by fold: https://www.cathdb.info/ So the value from that is probably the most useful for you.

Crystallographic conditions

  • pH
  • Salinity
  • Dissolved Sugars
  • Reducing Agent Concentration

Your crystallising conditions do not mean much. They are a solution that is close to precipitating your protein but slowly enough for it to crystallise. A protein can crystallise under multiple different conditions and a crystal may be chosen simply because it is a lower index well in the Hampton screening conditions... These mean very very little. And unless you are planning on doing ML to determine the best screening condition these should be avoided.

Data from the PDB

You have probably noticed that the only stats PDB/PDBe give are crystallographic, like space group, unit cell dimensions, resolution, completeness in shells, Rfree, Rmerge etc. So nothing there is of much use... except for bound ligand. Often this is the cofactor, but often it's a cofactor analog or a product (as opposed to the substrate). But that information can come from Uniprot etc.

pH

A solution has a pH, a protein has a pI. At a pH equal to the pI it will be neutral —and less insoluble. ProtParam can calculate the pI from a linear sequence. However, surface charge totally shows different spots. APBS can map the solution to the Poisson-Boltzmann equation to your protein. There is a PyMOL wrapper, a Pyrosetta wrapper and more. But you still would need to some stats and 3D operations to convert it into a small vector of values —rototranslations and projects are not for the faint of heart or in a rush.

Melting temperature

BRENDA has lots of enzymatic data on enzymes including melting temperature or kinetics at different temperatures (which follow the Eyring equation (TST), until MMRT kicks in) which could be useful.

For a given protein you can calculate the relative folding ∆∆G (eg. with Rosetta Score), which is a proxy to melting temperature in a buffer it is happy in.

Despite paining me to say... some fanciful Poisson-Boltzmann map derived vector and ∆∆G are probably all you could get from analysing structures. Whereas the rest are from CATH classification and other databases that are not structure based.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.