SuMo: A Tool for Protein Function Inference Based on 3D Structures Comparisons - Identification of Ligand Binding Site and Protein-Protein Interaction Area - page 2

(Doppelt-Azeroual et al. Protein Sci 19:847-867, 2010). In order to give access to
the SuMo tool, we proposed a web server (Jambon et al. Bioinformatics 21:3929-
3930, 2005) reachable at http://sumo-pbil.ibcp.fr . This chapter will describe the
main rationale we initially took for designing the first release of SuMo. In addition,
we propose a completely new set of parameters best suitable for proteins and finally,
we illustrate its power with several biological examples. Two of them dealing with
serine proteases and lectins are given for a comparison purpose. The first two
examples illustrate the capability of SuMo to deal with completely opposite modes
of evolution i.e. convergence and divergence. A new biological application dealing
with betalactame binding protein PBB molecules is also presented.

Keywords Proteins • Structural bioinformatics • 3D structure • Physico-chemical
groups • 3D sites • Annotation • Triangle form • Delta-plus • Delta-minus • Glycine
polar • Hydrophobic aliphatic • Carbon alpha • Hydrophobic aromatic • Target structure
• Objects • Proteases • Isomerases • Lectins • Betalactam • Penicillin drug • Cephalosporin
drug • Ceftazidime • Serine proteases • Protein-protein interaction • SuMo

1.1 Introduction

Understanding and predicting the function of proteins using bioinformatics traditionally
falls into three levels of knowledge: amino acid sequence, backbone structure (also
called fold comparison/recognition) and local arrangement of atoms (sites detection).
At the sequence level, similarity based methods such as FASTA (Pearson 1991 ) or
BLAST (Altschul et al. 1997 ) are commonly used by molecular biologists for
the retrieval of similar (or homologous) sequences. These methods are suitable for
finding homologous proteins that share similar folds. Still at the sequence level,
other tools exist that used the recognition of given patterns into protein sequences
(Hulo et al. 2008 ; Sigrist et al. 2010 ) . These methods can be used in non-homologous
proteins as their sequences can share common similar sub sequences exhibiting
common functions without the necessity to be homologues. However, receptor-ligand
interaction can be conserved in functionally equivalent proteins even in the absence
of sequence homology. If the 3D structure is available, the backbone level based
comparison methods mainly rely on RMSD calculation after structure superimposition
( Holm and Sander 1997 ). Alternatively, other methods rely on surface alignment
(Guerra et al. 2010 ) or surface matching (Via et al. 2000 ) algorithms. Although very
useful for a global fold comparison, this superimposition strategy is not suitable for
sites detection in arbitrary 3D protein structures (Jambon et al. 2003 ) . Although of
large interest in drug design or in deciphering function from 3D structures, methods
based on 3D sites prediction have been lately developed. Most of them have been
designed for the recognition of protein-ligand binding sites and the comparison of
protein-protein interfaces or protein pockets (Reisen et al. 2010 ) . They include hashing
techniques (Wallace et al. 1997 ; Shulman-Peleg et al. 2004 ) , evolutionary trace