search

rcsb / mmtf

The specification of the MMTF format for biological structures
http://mmtf.rcsb.org/
44 stars 17 forks source link

DSSP needs upgrade to 2.0 #3

Open BobHanson opened 8 years ago

BobHanson commented 8 years ago

The DSSP analysis that RCSB is using (from BioJava) is using an outdated DSSP algorithm. see Int. J. Mol. Sci. 2014, 15, 7841-7864; doi:10.3390/ijms15057841 for a discussion of DSSP 2.0. Jmol is moving to this default, so "DSSP" in Jmol will now point to this.

Suggestion is that RCSB do the same and incorporate DSSP 2.0, not DSSP 1.0, into mmtf.

Bob

andreasprlic commented 8 years ago

There are actually a number of problems with DSSP. If we deviate from the well known Kabsch & Sander DSSP, I believe we can actually do a number of things. There is a BioJava thread https://github.com/biojava/biojava/issues/454 that discusses improvements to the secondary structure implementation.

arose commented 8 years ago

From http://swift.cmbi.ru.nl/gv/dssp/index.html

In 2012 the determination of π-helices (I prefer to call them α-bulges) has been implemented in DSSP strictly following the original description by Kabsch and Sander, but assigning π-helices before α-helices. Output from the new version (2.1.0) therefore deviates a bit from output from the original DSSP (that now is called DSSPold). We believe that the newest DSSP version is ′better′ than all previous versions.

So, DSSP 2.0 is a very small incremental change as as I can see. What is the rationale to change the order of assignment, that is "π-helices before α-helices"?

The MMTF specification allows any per-residue secondary structure data, not just DSSP. There will be a page that exactly describes what the MMTF files we provide contain.

BobHanson commented 8 years ago

I have switched Jmol to DSSP 2.0. The rationale is that helices with "kinks" or "bulges" were being treated as one long helix. In many case, particularly in G-protein-coupled receptors, this results in situations where, especially with rocket rendering, a rendering of the helix does not follow the helix faithfully. DSSP 2.0 better represents that broken helix as three -- alpha-pi-alpha. The original DSSP almost never marks a pi-helix because it is generally just such a bulge. I spotted it with 1F88, but there are many other cases mentioned in that article.

What I've done in Jmol is to make the default to ignore the mmtf assignment and just do the DSSP 2.0 assignment itself. But if one really wants the older assignment, then what you can do is to add FILTER "DSSP1" to the load line. Then two things will happen: (1) The DSSP assignment in the mmtf file will be read, and (2) Jmol will not fill in missing models (as in 1CDR) with its own DSSP 2.0 calculation.

I suggest that if you are going to use DSSP at all, you should implement the upgrade to DSSP 2.0, as that is what the authors certainly want "DSSP" to mean now.

arose commented 8 years ago

DSSP 2.0 better represents that broken helix as three -- alpha-pi-alpha.

I see, that makes perfect sense. Thanks!

michel4j commented 7 years ago

@arose

So, DSSP 2.0 is a very small incremental change as as I can see. What is the rationale to change the order of assignment, that is "π-helices before α-helices"?

I wrote a paper about this a long time ago.

https://academic.oup.com/peds/article/15/5/353/1508254/Occurrence-conformational-features-and-amino-acid

DSSP uses hydrogen bonds to assign helices (1-->4 for α), (1-->5 π). The problem is that often both hydrogen bonds are present bifurcated and the order of assignment will impact the assignment severely. The change I suggested back then was to use hydrogen bond energy preference. The original DSSP was biased against π by assigning α first, the DSSP 2.0 is perhaps now biased against α. Why not use the hydrogen bond energy preference as I suggested?