||January 18, 2008
||Oliver Sander, Max-Planck-Institut-Informatik
||Inferring Molecular Function from Local Function Conservation in
Sequence and Structure Space
|| Inference of protein function based on similarity assumes that
function is conserved among similar proteins.
But the extent of function conservation may differ considerably
between different regions of protein
sequence and structure space.|
We present a new protein function prediction method Godot, which
determines local function conservation
among similar proteins. The method assesses the variability of
protein function in protein sequence and
structure space and annotates the query protein based on a metric
derived from this analysis. The degree
of function conservation is regarded as a confidence measure for
prediction, high conservation implying high
confidence that the respective function is correct.
This quantitative estimate yields a differentiated view on function
conservation, enabling us to predict protein
molecular function more accurately than standard methods. Furthermore,
we apply the method to 500
functionally unannotated PDB structures and discuss selected examples.