Information Technology Reference
In-Depth Information
Re
nement algorithms mostly don
'
t fold a target structure to its possible native
state. Model re
nement is still obstructed with incorrect energy function, integrated
with an additional complication of erroneous conformational search programs.
Model selection among hundreds of models generated by Modeller is still a
challenging task. However, these issues have been solved to some extent by
evaluating these models by various scores e.g. GDT-TS and TM Score etc.
Improvement in the current algorithms is needed for the selection of the best model
since till date there is no set benchmark for selection of the best model, even by top
ranked servers as per CASP.
9 Conclusion
Correct prediction of secondary structure is the key to predict a good or satisfactory
tertiary structure of the protein. Secondary structure not only helps in predicting the
tertiary structure but also helps in predicting the function as well as sub-cellular
localization of proteins. Staring from the amino acid propensity based secondary
structure prediction methods, machine learning approaches has revolutionized the
prediction accuracy of secondary structure from 60 to 80 %.
Tertiary structure prediction by bioinformatics or computational biology tools is
always a challenging task for scientists. Ab initio folding and threading are com-
putationally expensive methods for tertiary structure prediction which, also results
in protein structural models having low accuracy. Tertiary structure prediction by
ab initio folding/modelling still has a limitation due to searching a large number of
conformations generated as well as absence of suitable potential functions as the
number of amino acid increases. Another method is fold recognition where, the
prediction accuracy is better than ab initio folding/modeling. Homology modeling,
the third prediction method, has emerged as the sole method which can build the
model close to X-ray crystal/NMR structure. Therefore, among the three methods,
comparative or homology modeling is considered as the best method for protein
structure predication with high accuracy in such cases where the sequence identity
between the target and template sequence is more than 30 %. These comparative
models may be used for structure based drug designing as well as virtual screening
to identify novel inhibitors. Selecting the best model in homology modelling is one
of the major challenging tasks to look into. In homology modeling, the major
chances of error may be in loop modeling if long loop is present in the target protein
molecule. Side chain modeling is another challenging area where prediction
accuracy should be increased. Now a day, hybrid methods became popular because
they club together a number of features such as structural alignments, solvent
accessibility and secondary structure information in order to produce a protein
model with high accuracy. Along with hybrid methods, several meta-servers are
also available which integrate protein structure predictions performed by various
methods that assemble and interpret the results to come up with a consensus model
prediction. Nevertheless, we have not reached the pinnacle of that modelling
Search WWH ::
Custom Search