neural nets and protein structure prediction

Evan W. Steeg steeg at cs.toronto.edu
Wed Apr 15 11:33:21 EST 1992


In article <1992Apr14.184510.9403 at dartvax.dartmouth.edu> robert.h.gross at dartmouth.edu (Robert H. Gross) writes:
>Hi folks,
>
>I am interested in identifying some recent papers that address the use
>of neural nets in determining protein structure. Can someone please
>post some references?
>
>Thanks.
>
>Bob Gross
>Dartmouth College
>bob.gross at dartmouth.edu


 Here are the ones I have in my bibtex bibliography file right now:

This first one is the *first* paper on the topic.  Qian and Sejnowski
used a backprop net and NetTalk-like "local window" approach to
predict secondary structure.

@article{qian-sejnowski-88,
        title="Predicting the secondary structure of globular proteins using neura
l network models",
        author="Qian, N. and Sejnowski, T. J.",
        year="1988",
        journal="Journal of Molecular Biology",
        volume="202",
        pages="865-884",
        annote="
The best existing method for predicting the secondary structure
of a globular protein is a neural network:
Results also indicate that only marginal improvements
on our performance will be possible with local methods.
Tertiary (3-D) structure is a much more difficult problem
for which there are no good methods.
-- Terry"
        }


 The next few are also *secondary* structure prediction papers, and
offer some slight changes and extensions to the Qian/Sejnowski work.



@article        (holley-karplus-89,
Key     =       "Holley",
author  =       "Holley, L. H. and Karplus, M.",
title   =       "Protein Secondary Structure Prediction with a Neural Network",
year    =       "1989",
journal =       "Proceedings of the National Academy of Science USA",
volume  =       "86",
pages   =       "152-156"
)


@techreport     ( fogelman-90,
key     =       "Fogelman-Soulie" ,
author  =       "Mejia, C. and Fogelman-Soulie, F." ,
title   =       "Incorporating Knowledge in Multilayer Networks: The Example
                 of Proteins Secondary Structure Prediction",
type    =       "Rapport de Recherche" ,
institution=    "Laboratoire de Recherche en Informatique, Univ. de Paris-Sud" ,
year    =       "1990"
)


@article        (bohr-et-al-88,
Key     =       "Bohr",
author  =       "Bohr, H. and Bohr, J. and Brunak, S. and Cotterill, R. M. and Lau
trup, B. and Norskov, L. and Olsen, O. H. and Petersen, S. B.",
title   =       "Protein Secondary Structure and Homology by Neural Networks: The
Alpha-Helices in Rhodopsin",
year    =       "1988",
journal =       "FEBS Letters",
volume  =       "241",
pages   =       "223-228"
)

@article{Kneller-at-al-90,
author = "D. G. Kneller
  and F. E. Cohen
  and R. Langridge",
title = "Improvements in protein secondary structure prediction by an
  enhanced neural network",
journal = "J. Mol. Biol.",
volume = "214",
pages = "171-182",
year = "1990"}


 The group listed below came up with a very interesting result:  
A simple Bayesian one-shot classifier, embodying the *very* 
unrealistic assumption of statistical independence between adjacent
and nearby amino acid residues, worked nearly as well as the most
sophisticated methods.  This, I think, tells us quite forcefully
that the 60-70% predictive accuracy rate "wall" will never be
breached without discarding the local window approach and trying to
incorporate more global, and tertiary, structural information into
secondary structure prediction.

@techreport     ( lapedes-prot,
key     =       "lapedes",
author  =       "Stolorz, P. and Lapedes, A.~S. and Xia, Y",
title   =       "Predicting Protein Secondary Structure Using Neural
                 Net and Statistical Methods",
type    =       "Technical Report",
number  =       "LA-UR-91-15",
institution=    "Los Alamos National Laboratory",
year    =       "1991",
bibdate =       "Tue Mar  1 19:21:22 1988" ,
keywords=       "nonlinear signal processing, backpropagation"
)

 The Shavlik group at Wisc. has been working on some interesting
ways to move between neural network and rule-based approaches
to adaptive pattern recognition.  They have applied their methods
to several tasks in computational molecular biology, including
promoter recognition and protein secondary structure prediction.
Unfortunately, I seem to have lost 2 of the references, but here's
one:

@unpublished      (towell-shavlik-kbann,
   AUTHOR = "Towell, G.~G. and Shavlik, J.~W.",
   YEAR = "1991",
   TITLE = "The Extraction of Refined Rules from Knowledge-Based
            Neural Networks",
   NOTE = "Submitted to {\em Machine Learning}"
   )


The next several are about *tertiary* structure prediction.
The Wolynes paper below is not about neural nets per se, but describes
a NN-like method for classifying/recognizing/predicting protein structures.
(The Wolynes group has a new, 1992, paper out as well, but I don't
have the reference handy).


@article        ( wolynes-89,
key     =       "Wolynes" ,
title   =        "Toward Protein Tertiary Structure Recognition by
                  Associative Memory Hamiltonians",
author  =       "Friedrichs, M. S. and Wolynes, P. G." ,
journal =       "Science" ,
year    =       "1989" ,
volume  =       "246" ,
pages   =       "371-373"
)


@article        (bohr-et-al-90,
Key     =       "Lautrup",
author  =       "Bohr, H. and Bohr, J. and Brunak, S. and Cotterill, R. M. J. and
Fredhom, H. and  Lautrup, B. and Petersen, S. B.",
title   =       "A Novel Approach to Prediction of the 3-Dimensional Structures of
 Protein Backbones by Neural Networks",
year    =       "1990",
journal =       "FEBS Letters",
volume  =       "261",
pages   =       "43-46"
)


 Our work, described in the papers referenced below, combines
partial tertiary structure prediction with classification,
and is an attempt at combining bottom-up 
(sequence -> 2-ary struc -> 3-ary struc) information with top-down
(3-ary struc -> secondary struc) information in a unified global
constraint satisfaction approach.

@INPROCEEDINGS{greller-steeg-salemme-91a,
     AUTHOR = {Greller, L. D. and Steeg, E. W. and Salemme, F. R.},
     TITLE = {Neural Networks for the Detection and Prediction of 3D Structural Do
mains in Proteins},
     BOOKTITLE = {Proceeedings Eighth International Conference on Mathematical and
 Computer Modelling},
     YEAR = {1991},
     ADDRESS = {College Park, Maryland},
     MONTH = {April},
     ANNOTE = {A newer version has been submitted to {\em Science}.}
}


  I also review some of these approaches and the computational issues
in the following (which focuses on RNA structure, however).  There is
at least one other chapter in the book that describes protein structure
prediction with NNs, but, again, I don't have the names.  Sorry.

@incollection   ( steeg-bookchapter,
author    =     "Steeg, E.~W.",
year      =     "1992",
title     =   "Neural Networks, Adaptive Optimization, and
               {RNA} Secondary Structure Prediction",
booktitle =   "Artificial Intelligence and Molecular Biology",
editors   =   "Hunter, L.",
publisher =   "AAAI Press",
note      =    "In Press"
)



  That's all for now.

    -- Evan


-- 

Evan W. Steeg (416) 978-5182      steeg at ai.toronto.edu (CSnet,UUCP,Bitnet)
Dept of Computer Science          steeg at ai.utoronto    (other Bitnet)
University of Toronto,            steeg at ai.toronto.cdn (EAN X.400)
Toronto, Canada M5S 1A4           {seismo,watmath}!ai.toronto.edu!steeg



More information about the Comp-bio mailing list