Reader Comments
Post a new comment on this article
Post Your Discussion Comment
Please follow our guidelines for comments and review our competing interests policy. Comments that do not conform to our guidelines will be promptly removed and the user account disabled. The following must be avoided:
- Remarks that could be interpreted as allegations of misconduct
- Unsupported assertions or statements
- Inflammatory or insulting language
Thank You!
Thank you for taking the time to flag this posting; we review flagged postings on a regular basis.
closeTorusDBN by Boomsma et al.
Posted by thamelry on 20 May 2010 at 11:41 GMT
Dear colleagues,
The statistically rigorous description of biomolecular structure is an important and timely endeavor, which promises to revolutionize protein structure prediction, simulation and design. We compliment the authors for a contribution to an important emerging field.
However, we would also like to point to a closely related article that was published by us in 2008 in Proc. Natl. Acad. Sci. USA. <http://www.pnas.org/conte...>. This article has unfortunately escaped the attention of Ting et al., it seems.
Our article also presents a probabilistic model (called TorusDBN) of the phi, psi backbone angles in continuous space. As suggested (but not implemented) by Ting et al., our work combines a graphical model with directional statistics, resulting in a probabilistic model that incorporates long-range neighbor dependencies.
As the authors rightfully point out, "Many other such statistical analyses [of] phi,psi data have appeared previously [...], but few are publicly available for use in structure validation or structure prediction." We point out that our method is freely available from SourceForge <http://sourceforge.net/pr...> under the GPL license.
Best regards,
Wouter Boomsma, University of Cambridge, UK
Kanti V. Mardia and Charles C. Taylor, University of Leeds, UK
Jesper Ferkinghoff-Borg, DTU, Denmark
Anders Krogh and Thomas Hamelryck, University of Copenhagen, Denmark
RE: TorusDBN by Boomsma et al.
dunbrack replied to thamelry on 20 May 2010 at 22:42 GMT
Thanks for pointing out your paper, Thomas. And I apologize that we neglected to cite it in our paper.
I guess the two methods may be used in different ways, or at least in different contexts of how people construct their structure prediction programs. Ours is a set of files that contain probability distributions, which people may use however they wish and they may find them easy to understand and use. At least some of the neighbor interactions that produce different effects may be readily observed.
I gather yours is a program that generates probabilities or samples on demand, and as with hidden Markov models it is not necessarily designed for interpretation of the transition probabilities or emission probabilities. However it may be better at generating samples and clearly has some features not included in ours (secondary structure effects, for instance; ours is based only on loop conformations, and in particular optimized for loop structure prediction).