Mm. Gromiha et al., Role of structural and sequence information in the prediction of protein stability changes: comparison between buried and partially buried mutations, PROTEIN ENG, 12(7), 1999, pp. 549-555
Predicting mutation-induced changes in protein stability is one of the grea
test challenges in molecular biology, In this work, we analyzed the correla
tion between stability changes caused by buried and partially buried mutati
ons and changes in 48 physicochemical, energetic and conformational propert
ies. We found that properties reflecting hydrophobicity strongly correlated
with stability of buried mutations, and there was a direct relation betwee
n the property values and the number of carbon atoms. Classification of mut
ations based on their location within helix, strand, turn or coil segments
improved the correlation of mutations with stability. Buried mutations with
in beta-strand segments correlated better than did those in alpha-helical s
egments, suggesting stronger hydrophobicity of the beta-strands. The stabil
ity changes caused by partially buried mutations in ordered structures (hel
ix, strand and turn) correlated most strongly and were mainly governed by h
ydrophobicity, Due to the disordered nature of coils, the mechanism underly
ing their stability differed from that of the other secondary structures: t
he stability changes due to mutations within the coil were mainly influence
d by the effects of entropy. Further classification of mutations within coi
ls, based on their hydrogen-bond forming capability, led to much stronger c
orrelations. Hydrophobicity was the major factor in determining the stabili
ty of buried mutations, whereas hydrogen bonds, other polar interactions an
d hydrophobic interactions were all important determinants of the stability
of partially buried mutations. Information about local sequence and struct
ural effects were more important for the prediction of stability changes ca
used by partially buried mutations than for buried mutations; they strength
ened correlations by an average of 27% among all data sets.