Secondary and Tertiary Structure Prediction of Proteins: A Bioinformatic Approach - Complex System Modelling and Control Through Intelligent Soft Computations - page 535

methodologies or algorithms as well as key steps which need to be improved are
discussed in Sect. 8 . Finally, Sect. 9 provides a comprehensive conclusion for the
entire chapter.

2 Secondary Structure Prediction

2.1 Amino Acid Propensity Based Prediction

Early prediction methods as proposed by Chou and Fasman ( 1974 ) and the Garnier-
Osguthorpe-Robson (GOR) (Garnier et al. 1978 ) rely on the propensity of amino
acids that belong to a given secondary structure. These are simple and direct
methods, devoid of complex computer calculations, that utilize empirical rules for
predicting the initiation and termination of helical regions in proteins. The relative
frequencies of each amino acid in each secondary structure of known protein
structures are used to extract the propensity of the appearance of each amino acid in
each secondary structure type. Propensities are then used to predict the probability
that amino acids from the protein sequence would form a helix, a beta strand, or a
turn in a protein. These methods have introduced the conditional probability of
immediate neighbor residues for computation. The web-servers based on Chou and
Fasman ( 1974 ) and GOR showed prediction accuracy between 60

65 %. However
the updated, GOR V algorithm which is available as web-server at http://gor.bb.
iastate.edu/ combines information theory, bayesian statistics and evolutionary
information and has reached an accuracy of prediction to 73.5 % (Sen et al. 2005 ).

-

2.2 Template Based Prediction

This method uses the information from database of proteins with known secondary
structures to predict the secondary structure of a query protein by aligning the
database sequence with the query sequence and finally assigning the secondary

structures to the query sequence. The nearest-neighbor method belongs to this
category. This category is reliable if both sequences have good identical or
homologous regions as compared to a threshold value. The two most successful
template-based methods are Nearest-neighbor Secondary Structure Prediction
(NNSSP) (Yi and Lander 1993 ) and PREDATOR (Frishman and Argos 1997 ). The
accuracy of these methods lies in the range 63 68 % (Runthala and Chowdhury

-

2013 ).